Understanding the Development and Design of Chinese Cities: Towards an Approach Based Upon the New Science for Cities

Understanding the Development and Design of Chinese Cities: Towards an Approach based upon the New Science for Cities

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

Hui Kong

Graduate Program in Geography

The Ohio State University

2018

Dissertation Committee:

Daniel Sui, Advisor

Morton O’Kelly

Ningchuan Xiao

Max Woodworth

Copyrighted by

Hui Kong

2018

Abstract

Cities are complex self-organizing systems. A holistic urban design needs to take into consideration all the sub-systems, driving forces, and stakeholders as well as the interactions among them. For a long time, traditional urban planners and designers in

China have to work under an over-simplified conceptual framework and methodology to produce scenarios/policies that influence the life of citizens. Recently, with the explosion of big data and high-performance computing, the emerging new urban science has opened up important avenues to better understand the complexity of cities, thus to support urban design practices. Quantitative methods such as spatial data analysis and urban models have become the essential tool and also posed critical challenges for the design and planning of Chinese cities under the new data environment. During the past decade, numerous quantitative urban studies have been conducted considering urban modeling or urban planning of Chinese cities. However, most studies have failed to synergistically integrate quantitative methods with design practices. To fill in the gap between urban design and quantitative urban studies, this dissertation aims to build a bridge between urban planning/design and quantitative geographic data analysis. The new science of cities is introduced as the theoretical foundation of this research. In the new science of cities, the quantitative methods (positive/empirical dimension of urban studies) and urban design (normative/idealistic dimension of urban studies) are seamlessly integrated to help us understand and design cities (Batty, 2013). Inspired by ii the framework as articulated in the new science of cities, this doctoral dissertation aims to develop a synthetic approach to link the positive and normative dimensions, to accomplish urban design based on spatial data analysis and urban simulation. By doing so, we develop new approaches for understanding and designing of Chinese cities, from three essential geographical perspectives in the context of new city science -space, place, and network. Case studies are conducted from each perspective to test the potential applications of the proposed synthetic approach. Urban areas in Changping (part of

Beijing) and Shenzhen of China are chosen as the case study sites, to conduct urban designs with specific policy goals: controlling urban sprawl, promoting mixed-use development, and improving transport network accessibility. Quantitative data analysis and urban simulation are applied to the urban design process in this study.

iii

Acknowledgments

Over the past five years, I have been through a wonderful journey full of challenges, unexpectedness and sweetness. There are so many people I am thankful of, and I would like to present my sincerest gratitude to the following persons who have kindly offered me the support, accompany, and encouragement during the journey of completing my Ph.D. degree.

First and foremost, I want to thank my advisor, Dr. Daniel Sui, for his persistent support in the past five years. I was so nervous when first coming to the US and entering the realm of research. It is Dr. Sui who patiently led me into the academic sphere and helped me establish my own style of doing research. Dr. Sui is very motivating, enlightening, and supportive during my pursuit of the degree. He is the one whom I can also turn for help, and is always so supportive and encouraging to both my professional and personal growth. His faith in me enables me to hang on even in my hardest time, and his passions and seriousness on academia set a good example to me. It is my fortune to have Dr. Sui as my advisor, and his intellectual inspiration will have a life-long influence on my academic career.

My sincere appreciation also goes to my dissertation committee, Dr. Morton

O’Kelly, Dr. Ningchuan Xiao, and Dr. Max Woodworth, for their time and effort in

iv refining my research design and improving my manuscript. I could not finish my dissertation without their informative input and careful feedback.

To Scarlett Jin, you are like my sister. I have always been able to turn to you when I need a sympathetic ear and an honest opinion. The collaboration with you is so productive and delightful. I will miss the days when we study together either in our office or in the library. To Xining Yang and Bo Zhao, thanks for being a good example of mine and for your kind help and advice during my pursuit of Ph.D. To Ning Zhang, Jay Liang,

Chen Zhao, Zhiqi Yu, Yuechun Wang, and Yapin Liu, your company has made my time at OSU the best time in my life. And to Dr. Yang Yue, Dr. Yongxi Gong, Dr. Jincheng

Jiang, and Qili Gao, thanks for your support of my research during my stay in Shenzhen last summer.

Last but not least, I want to thank my family. I am particularly grateful to my parents and in-laws, who have always inspired me to follow my heart and supported me to pursue my dreams. Thank you, my one and only, Zhenong Jin. You are always there for me, being my persistent source of strength, joy, and love. Without your companion, encouragement, and help, this dissertation will never be able to see the light of the day.

Being able to meet you and marry you, is the luckiest thing in my life!

Vita

2009...... Zhangzhou No.1 High School,

Zhangzhou, China

2013...... B.S. Urban & Rural Planning,

Peking University, China

2015...... M.A. Geography, The Ohio State University

2015 to present ...... Graduate Teaching Associate, Department

of Geography, The Ohio State University

Publications

Gao, Q. L., Li, Q. Q., Yue, Y., Zhuang, Y., Chen, Z. P., & Kong, H. (2018). Exploring

Changes in the Spatial Distribution of the Low-to-Moderate Income Group using

Transit Smart Card Data. Computers, Environment and Urban Systems. Accepted.

Jin, S. T., Kong, H., Wu, R., & Sui, D. Z. (2018). Ridesourcing, the sharing economy,

and the future of cities. Cities.

Tao, Z., Zheng, Q., & Kong, H. (2018). A modified gravity p-median model for

optimizing facility locations. Journal of Systems Science and Information.

Accepted.

Sui, D. Z., Zhao, B., & Kong, H. (2017). The Development of Copycat Towns in China:

An initial analysis of their economic, social, and environmental implications.

Boston, MA.: Lincoln Institute. Entire report available on-line at:

http://www.lincolninst.edu/publications/working-papers/development-copycat-

towns-china

Kong, H., & Sui, D. Z. (2017). Integrating the normative with the positive dimension of

the new science for cities: A geodesign-based framework for Cellular Automata

modeling. Environment and Planning B: Urban Analytics and City

Science, 44(5), 837-863.

Kong, H., Sui, D. Z., Tong, X., & Wang, X. (2015). Paths to mixed-use development: A

case study of Southern Changping in Beijing, China. Cities, 44, 94-103.

Wang, X., Tong, X. & Kong, H. (2013). Occupational Flows and Structural Change: an

Empirical Study in Northern Beijing. Special Zone Economy. (6), 28-32 (In

Chinese)

Kong, H., Jin, S. T., & Sui, D. Z. Uber, public transit, and transportation equity.

Professional Geographer. (under review)

Tao, Z., Yao. Z., Kong, H., Duan, F., & Li, G. Measuring healthcare accessibility using

the multi-modal two-step floating catchment area method in Shenzhen, China:

estimating travel time via online map APIs. BMC Health Services Research.

(under review)

Fields of Study

Major Field: Geography vii

Table of Contents

Abstract ...... ii

Acknowledgments...... iv

Vita ...... vi

Table of Contents ...... viii

List of Tables ...... xi

List of Figures ...... xii

Chapter 1: Introduction ...... 1

1.1 Motivation and Context ...... 1

1.2 Research Objectives ...... 3

1.3 Synopsis of the Dissertation ...... 5

Chapter 2: Research Background and Conceptual Framework ...... 8

2.1 Urban Planning in China ...... 8

2.2 Data analytics and its applications ...... 13

2.3 Positive and Normative Dimensions of City Science ...... 22

2.4 Theoretical Framework Linking Spatial Data Analysis with Urban Planning ...... 23

Chapter 3: Space Perspective - Integrating CA Modeling with Geodesign ...... 26

viii

3.1 Research Context...... 27

3.2 Geodesign-based CA Modeling Framework ...... 31

3.3 Data and Methodology ...... 35

3.4 Results ...... 46

3.5 Discussions and Conclusions ...... 54

Chapter 4: Place Perspective - Integrating Questionnaire Analysis with Mixed-use

Development ...... 58

4.1 Research Context...... 59

4.2 Alternative Paths to Mixed-use Development ...... 62

4.3 Study Area, Data, and Methodology ...... 67

4.4 Results ...... 77

4.5 Discussions and Conclusions ...... 92

Chapter 5: Network Perspective - Integrating SCD Analysis with Transportation Design

...... 95

5.1 Research Context...... 95

5.2 Transportation Design: Gap and Redundancy ...... 99

5.3 Study Area, Data, and Methodology ...... 106

5.4 Results ...... 114

5.5 Discussions and Conclusions ...... 136

Chapter 6: Summary and Conclusion ...... 138

6.1 Summary and Conclusions ...... 138

6.2 Future Research ...... 141

References ...... 144

List of Tables

Table 1. Driving Factors of Urban Development in Changping District ...... 39

Table 2. Prediction results of urban development in 2020 ...... 48

Table 3. Land-use Impact Indicators of six scenarios (smaller number indicates better performance) ...... 50

Table 4. LIIs of new design and six previous design scenarios ...... 54

Table 5. Three types of communities and development models ...... 70

Table 6. Build environment of three communities ...... 75

Table 7. Comparison of workplace-residence separation rate ...... 78

Table 8. Results of homogeneity test of variances in the sense of community analysis... 81

Table 9. One-way ANOVA analyzing results of the sense of community ...... 83

Table 10. Results of homogeneity test of variances in community vitality analysis ...... 85

Table 11. One-way ANOVA analyzing results of the community vitality in community 86

Table 12. Comparing results of three communities ...... 87

Table 13. Supply criteria and indicators ...... 104

Table 14. Improvement of the bus service supply after optimization ...... 134

List of Figures

Figure 1. The two-tier system of urban planning in China [Source: modified from Wu,

2015] ...... 11

Figure 2. GIS Application to Different Functions of Urban Planning ...... 15

Figure 3. Theoretical Framework: Integrating the Two Dimensions of City Science ...... 24

Figure 4. The Framework of Geodesign [source: Steinitz (2012)] ...... 32

Figure 5. Conceptual framework: integrating CA modeling with geodesign ...... 34

Figure 6. Study area: Changping district in Beijing, China ...... 36

Figure 7. Urban growth of Changping district from 2005 to 2012 ...... 37

Figure 8. Six design scenarios of the study area ...... 38

Figure 9. Goodness-of-fit (GOF) of all experiments...... 47

Figure 10. Land-use impact indices (LIIs) of the six scenarios at the village level ...... 51

Figure 11. A new geodesign scenario for Changping ...... 52

Figure 12. Simulation of urban growth in 2020 under the new design scenario ...... 53

Figure 13. The iterative framework of geodesign-based CA modeling for urban growth design...... 56

Figure 14. Development process, urban form, and urban performance ...... 67

Figure 15. Study area and case study spots...... 69

Figure 16. Communities in Large-scale Residential Clusters (LRC)...... 71

Figure 17. Urban Villages...... 72 xii

Figure 18. Village-collective Agent Community (VAC)...... 74

Figure 19. The comparison of residents’ annual income...... 80

Figure 20. The radar chart of the three types of communities ...... 87

Figure 21. The village-enterprise joint development of Zhenggezhuang village ...... 91

Figure 22. Heuristic algorithm of bus stop location optimization ...... 113

Figure 23. Data processing and OD matrix estimation of bus SCD ...... 117

Figure 24. Space match: (a) space match by comparing boarding time with GPS time (b) an example of space match ...... 118

Figure 25. Matching bus routes and bus lines through inverse selection by probability (a) bus line match between two datasets (b) an example of inverse selection by probability

...... 119

Figure 26. Orientation extraction methods ...... 120

Figure 27. Alighting inference based on transition rules ...... 121

Figure 28. Supply of the bus services in Shenzhen from the access (top) and efficiency

(middle) perspective, and the overall supply measure (bottom)...... 126

Figure 29. Travel demand of the citizens in Shenzhen from the bus SCD (top) and taxi trip data (middle), and the overall measure of demand (bottom)...... 128

Figure 30. Gaps of bus supply in Shenzhen (top) and the stripes with concentrated gaps

(bottom)...... 129

Figure 31. Bus supply redundancy in Shenzhen ...... 130

Figure 32. Scenarios of the bus stops being built and removed...... 132

Figure 33. Design of bus stop locations ...... 133

xiii

Figure 34. The gap and redundancy of the new design scenario ...... 133

Figure 35. Change of gap as the number of stops being added ...... 134

Figure 36. Change of redundancy as the number of stops being removed ...... 135

xiv

Chapter 1: Introduction

1.1 Motivation and Context

Cities are complex self-organizing systems. Therefore, a holistic urban design needs to take into consideration all the sub-systems, driving forces, and stakeholders as well as the interactions among them for their successful planning and design. Due to the lack of knowledge about all the systems concerned, for a long time, traditional urban planners and designers in China have to work under an over-simplified conceptual framework and methodology to produce scenarios/policies that influence the life of citizens. Recently, with the explosion of big data and high-performance computing, the emerging new urban science has opened up important avenues to understand complex cities, thus to support urban design practices. Quantitative methods such as spatial data analysis and urban models have become the essential tool and also posed critical challenges for the design and planning of Chinese cities under the new data environment.

During the past decade, numerous quantitative urban studies have been conducted considering urban modeling or urban planning of Chinese cities. However, most studies concentrated on problem identification/evaluation of urban systems or empirical application of urban models, but failed to have a perspective on the application of quantitative methods to design practices. 1

To fill in this gap between urban design and quantitative urban studies, I will build a bridge between urban planning/design and quantitative geographic data analysis.

The new science of cities will be introduced as the theoretical foundation of this research.

In his recent work The New Science of Cities, Batty (2013) discussed the positive

(focusing on ‘what is’) dimension of the city science and the normative (focusing on

‘what should be’) aspect of it. In the field of urban studies, the positive approach, or empirical approach, could be related to quantitative urban studies, such as urban modeling and spatial data analysis; while the normative approach, or idealistic approach, is more concerned with the tradition of urban design. In the new science of cities, these two dimensions of city sciences could and should be better combined to examine urban issues.

Therefore, this doctoral dissertation will integrate the positive and normative dimensions of new city science, to accomplish urban design based on spatial data analysis and urban simulation. By doing so, I aim to develop new approaches for understanding and designing of Chinese cities, from three essential geographical perspectives in the context of new city science: space, place, and network. Case studies will be conducted on each perspective to verify the potential application of quantitative methods to urban design. Urban areas in Changping (part of Beijing) and Shenzhen of China will be investigated, to conduct urban designs with specific policy goals: controlling urban sprawl, promoting mixed-use development, and improving transport network accessibility. Quantitative data analysis and urban simulation will be applied to the urban design process in this study.

1.2 Research Objectives

Inspired by the dual dimensions of new city science, the overall goal of my doctoral dissertation is to integrate the positive and normative dimensions of city science, by combining urban simulation and quantitative analysis with urban planning/design.

Some specific Chinese cities would be selected to test this integration, because of the special development process and urban planning system in China. Firstly, after the Chinese Communist Revolution in 1949, most land is owned by collectivities or by the state. Therefore, the government has powerful control over the land use, construction and development. Under this situation, the urban design processed by government plays an essential role in the urbanization process. Secondly, the urban planning in China is a product of both top-down and bottom-up processes. After the economic reform in 1978, bottom-up urban planning has become more and more important in the planning system.

This trend enables different stakeholders to speak their voice and participate in the urban design process, makes urban design a more multi-agent negotiation process, as distinct from the traditional physical planning that is decided mainly by the ‘most authorized leader’ in the government. Under this situation, some new approaches should be put forward to replace the traditional physical design, to achieve a more reasonable and more

‘scientific’ urban design. Thirdly, many Chinese cities, especially large cities, are experiencing fast urbanization in the past two decades, and lots of urban problems are emerging. To respond timely and appropriately to the urban problems and fast-changing situations, urban design needs to be based on the comprehensive analysis and

3 understanding of the complex urban system. Therefore, quantitative urban studies would be a great tool to meet the new requirements of urban design.

Quantitative urban design from three perspectives – space, place, and network – is carried out. From each perspective, there will be different targets for urban design, which are either essential to healthy urbanization of high-density area, or are development goals set by local government.

From a space perspective, I will investigate the whole urban area of an administrative district. One essential goal of design in Chinese cities, considering the whole urban area, is to control urban sprawl so that the urban growth will occur at a reasonable speed, occupy less farmland while preserving the ecological system.

Therefore, the research questions on space aspect include: (1) what are the driving factors of fast urbanization? (2) Can existing urban design scenarios effectively control urban sprawl? (3) How to generate a better design scenario that controls urban sprawl, avoids occupation of essential farmlands and preserve ecological system?

From a place perspective, the design target is to achieve mixed-use development of communities in urban fringe areas, which will bring about compact form, walking- friendly built environment and vital community life to residents in the community. These research questions will be investigated: (1) with the strategy of mixed-use development being carried out, does the community realize the goal of mixed-use development? (2)

How to quantify the performance of a community considering mixed-use development?

(3) How to realize the real benefits of mixed-use development by planning and development?

From the network perspective, I will study aspects of the transportation network.

Using the bus system as an example, the gap and redundancy of the current system will be assessed, and the design of bus stops locations will be generated through an optimization process. I will answer the following questions: (1) what is the supply and demand of the current bus system? (2) Is there any gap and redundancy in the current bus service supply? How do the gap and redundancy distribute over space? (3) How to optimize the locations of bus stops to fill in the gap and reduce supply redundancy of the whole system?

Through the studies on the three perspectives, this dissertation research will be able to cover some of the deficiencies in the urban design of lacking geographic data platform and advanced spatial analysis, and thus help with the establishment of a more scientific quantitative urban design framework. Specifically, there are three primary research objectives:

(1) Design an urban growth scenario to control urban sprawl based on Cellular Automata

(CA) modeling;

(2) Investigate into alternative development processes to achieve mixed-use development of communities, to figure out the most effective path to this design target;

(3) Design an optimal bus system for less gap and redundancy based on bus smart card data analysis.

1.3 Synopsis of the Dissertation

This dissertation is organized as follows:

Chapter 1 is the introduction of the study, discussing the motivation and objectives of this research.

Chapter 2 reviews the historical change of urban planning in China, and the opportunities brought by the explosion of open/big data and data analytics. The gap between urban planning and data analytics in China is then discussed, followed by the theoretical framework of this dissertation that tries to fill in this gap. The theoretical framework aims at integrating the positive (‘what is’ issues) with the normative (‘what should be’ issues) dimensions in the new science of cities, and the Chapter 3 to 5 of the dissertation will test this theory from the space, place, and network perspectives.

Chapter 3 tests the integration of the positive and normative urban studies from the space perspective. In the case study conducted in Changping, Beijing, a geodesign- based Cellular Automata (CA) model is put forward to simulate and evaluate the existing urban design scenarios, and a new design scenario that can better control urban sprawl is generated based on the evaluation.

Chapter 4 explores the development process to achieve the claimed benefits of mixed-use development from the place perspective. Using southern Changping in Beijing as a case study, three typical urban development models commonly used in China – “top- down” centrally-controlled development model, “bottom-up” individual-dominant development model, and “bottom-up” collective-dominant development model – are compared and contrasted to figure out the effective development process to achieve mixed-used development.

Chapter 5 presents the integration of data analytics with urban planning from the network perspective. Bus smart card data and taxi trip data are analyzed to identify the gap and redundancy of the service supply of the current bus system in Shenzhen, China, and the design of bus stop locations is generated through an iterative optimization process to minimize the gap and redundancy of the whole system.

And finally, Chapter 6 is the conclusion of this dissertation. This chapter concludes this study, indicates its limitation, and provides potential directions for future research.

Chapter 2: Research Background and Conceptual Framework

2.1 Urban Planning in China

Since the establishment of the People’s Republic of China in 1949, China has been through dramatic transformations in both urban development and urban planning systems, as the consequence of political/economic environment change and technology advancement.

After the socialist revolution in 1949, Chinese cities have sustained low levels of urbanization in three decades, accompanied by the dominance of centralized, state- controlled urban planning and city-based industrialization (Ma, 2002). However, since the economic reforms in 1978, significant economic and spatial shifts away from the socialist patterns occurred in Chinese cities, followed by the accelerated urbanization, large-scale rural-urban migration, spatial reorganization through urban land-use change, urban housing markets development, and polycentric restructuring in urban form (Ma,

2002). At the same time, there are remarkable changes in urban planning and administrative systems. Traditional central-controlled urban planning was challenged by

8 the trends of decision-making decentralization, market-led urban development, socialist ideology retreat, and the increase in the number of actors and conflicts of interests in urban development and construction (Yeh & Wu, 1999). Urban planning practice in

China has shifted from centralized to decentralized, from top-down to bottom-up, and from passive static planning to more active dynamic planning. As the urban planning is no longer decided only by ‘one authorized person’, but a more collective process with a negotiation between several stakeholders, new approaches need to be taken into action instead of the traditional physical planning methods.

Historically, Chinese urban planning has been defined by two distinct traditions..

The first is the importance of cosmology as a guideline for houses and cities layout (Wu,

2015). The sites and layouts of cities and buildings are determined carefully based on cosmology, to serve as both as a defensive structure and a representation of ordered social structures (Wu & Gaubartz, 2013). Despite the traditional beliefs in cosmology, the modernist planning approach has been brought into China by bunches of architects and planners trained in the West (Wu, 2015). Not surprisingly, the second tradition emphasized the ‘scientific’ analysis and administrative efficiency in urban planning.

Therefore, this combination of ideological consideration as the cosmology traditions and the modernist planning approaches, makes urban planning in China a modernist process embedded in its special social and cultural ideals.

Since 1990, China has developed a two-tier statutory planning system (Figure 1).

The upper tier is the urban master plan (chengshi zongti guihua) that outlines the function, structure, population size and general land use of a city, and the lower tier is the

9 detailed plan (xiangxi guihua) that determines the detailed layout of construction to be carried out in the near future, including layout of transportation lines and facilities, characteristics of buildings, and other specific controls in the planning plot. In practice, there are two parts of the detailed plan: the detailed development control plan (DDCP)

(kongzhixing xiangxi guihua) and the detailed construction plan (DCP) (xiujianxing xiangxi guihua). DDCP is more regulatory, as it sets the land-use type, land-use intensity, and some other index to guide the DCP, while DCP is a more deterministic plan that designs the buildings, green space, transportation system and infrastructure facilities in the planning plot. Besides, there are some supplementary plans to support the two tiers.

For example, an urban system plan (chengshi tixi guihua) is sometimes prepared before the master plan, to streamline the development and hierarchy of cities and towns within the municipal area. Also, a strategic planning outline (guihua gangyao) is usually worked out immediately prior to urban master plan, to set the principles and major issues of the master plan, and provide some analysis on the technical, economic and political conditions of the planned area. For large cities, a district plan (fengqu guihua) may be prepared after a master plan to further control land uses at the district level and provide details for the detailed plan.

Figure 1. The two-tier system of urban planning in China [Source: modified from Wu,

2015]

With regard to the statutory planning system, there are some distinct characteristics of the urban planning in China. Firstly, the urban planning carried out by planners is separated from the politics in the policy-making of land development and economic planning commissions through the distribution and approval of construction projects. Especially with the decentralization of decision making and the trend of market- led development, more and more investment and construction initiatives are now coming from bottom-up, and urban planning in China sometimes fail to control and manage those bottom-up developments. Secondly, the mainstream planning process in China is still top-down and lacks public participation, which has led to difficulties in plan

11 implementation. Thirdly, the political interference in urban planning often occurs, when the administrators of government use their power to disturb the desired outcome of planning. This political interference will result in poor planning and decision making, poor implementation of the planning or unnecessarily prolonged planning process.

During the past decade, under the trend of big/open data explosion and smart city paradigm, urban planning of Chinese cities has come to another crossroad. The new data and technology environment provides us with opportunities to describe, analyze and understand our complex cities and urban spaces in a more elaborate way, while on the other hand raises challenges for urban planning/design on how to take advantage of the large-volume fine-scale data and advanced tools. Currently, numerous quantitative urban studies have been conducted utilizing urban big data and new techniques. However, most studies concentrated on the analysis, evaluation and problem identification of urban issues, while few of them have brought perspectives into urban planning and design.

Among the handful of research on quantitative urban design, Beijing City Lab

(BCL) has presented a new planning and design methodological paradigm termed Data

Augmented Design (DAD). DAD provides supporting tools covering the whole urban design process including investigation analysis, project design, evaluation and feedbacks based on quantitative urban analysis approaches such as data analyzing, modeling and forecasting (Long & Shen, 2015). Moreover, some scholars introduced the concept of

Planning Support System (PSS) developed by Harris (1960) into Chinese urban planning, and tried to advocate the application of planning software to Chinese urban planning

12 practice (Du & Li, 2005), establish techniques for PSS (Zhu & Shen, 2008; Long et al.,

2011), and put forward the framework of PSS (Long & Shen, 2015). These studies have preliminarily aroused public attention to PSS and provided some framework and techniques to achieve PSS in Chinese urban planning. However, more efforts need to be put on the empirical studies of planning support application in Chinese cities, the establishment of a more theoretical-based and applicable framework, and the integration of empirical quantitative urban studies and idealistic urban design. This research will fill in this gap by building the bridge that links urban design with quantitative urban studies such as geospatial analysis and urban modeling.

2.2 Data analytics and its applications

In retrospect, empirical urban studies have experienced a shift from qualitative to quantitative research: the urban studies began with the simple record and description of urban phenomena, and moved to the generalization of these phenomena, then to the analysis of relationships within urban systems, and finally to a systematic view of urban development (Liu et al., 2014). Under this trend, urban analytics has become the new hotspot in recent empirical studies of cities. Urban analytics refers to the scientific research approach that takes advantage of data and advanced techniques to investigate general urban rules, simulate urban growth, solve urban problems and evaluate urban policies. In this section, we review the data analytic approaches that are the focus of this dissertation and their application in urban planning. The approaches include GIS, urban models, and techniques related to big data.

2.2.1 GIS in urban planning

GIS is becoming increasingly accessible to urban planners and is now serving as both a spatial database and an analysis tool for urban planners in China. Database management, visualization, spatial analysis, and spatial modeling are the major application of GIS in urban planning (Fotheringham & Rogerson, 2013). How GIS intersects with urban planning varies according to the different functions, scales, and stages of urban planning.

With regard to the politics of China’s urban planning, there are mainly three functions of planning: general administration, development control and plan making

(Yeh, 1999). GIS intersects with all these three functions, but for each function, the application of GIS varies. We can conclude from Figure 2 that GIS serves as a database in general administration and development control of urban planning, and as a spatial analysis tool mainly in plan making. Also, the visualization function of GIS plays an important role in all three categories of urban planning functions.

120%

100%

80%

60%

40%

20%

0% Database Visualization Spatial Analysis Spatial Modeling Management

General Administration Development Control Plan Making

Figure 2. GIS Application to Different Functions of Urban Planning

[Source: Yeh, 1999 and generated by author]

Also, GIS techniques intersect with different tiers in the two-tier system of urban planning. Urban planning of each tier covers different scale of the planning area, ranging from the urban system, a whole city, to a district, or a street block in the city. When applying GIS to urban planning, raster data are more useful for large-scale planning, such as the city-wide strategic planning, where high resolution is not required, while vector data analysis is generally used for the detailed planning.

In the urban planning process, there are several stages for each tier in the system: the determination of planning objectives, the analysis of existing situations, modeling and projection, development of planning options, selection of planning options, plan implementation, and planning evaluation (Yeh, 1999). GIS can only provide data and techniques in some of the stages. For example, GIS has seldom been used in the

15 determination of planning objectives and plan implementation in China. The most important role GIS is playing is in the analysis of existing situation stage, by acting as the database storing the data for analysis and providing useful analysis tools such as map overlay, buffer analysis and spatial statistics. GIS can also be used for prediction and projection, bus due to some deficiencies of spatial models in urban planning practice, the urban planners in China has seldom used GIS in this stage. Land suitability maps generated in GIS are very useful in the development of planning options, and has been applied to many urban planning practices in China. The final planning selection of planning option in China is more a political rather than technical process, but planners can use GIS to generate some analytical results to provide some scientific suggestions to decision makers.

With regard to different sectors of urban planning, GIS intersects most with the land-use, transport, housing and environmental planning (Yeh, 1999). Key examples include site selection and land suitability analysis in land-use and housing planning, network analysis and route selection in transport planning, and buffer analysis and overlay processing in environmental planning.

2.2.2 Urban models in urban planning

Urban models have seldom been applied to urban planning in China, due to several reasons. Firstly, urban planning in China has a long tradition of physical planning, so the planners and policy makers did not have sufficient knowledge and appeal for models. Second, there are some structural problems exist in most urban models. Most

16 models were predicting the wrong sorts of thing such as long-term structural changes, which was not really what planners and policymakers were interested in. Third, most urban issues that urban planners need to consider are too complicated to model, and many models fail to grasp the most important elements of policy making.

Luckily, with improvement in related computational techniques, and also with the trend of big data and smart city, urban models are improving themselves. Also, urban planners are becoming more adaptable to using models in urban planning. Urban models are especially useful in the analysis process for strategic planning, and the evaluation of different planning scenarios. Also, some land-use simulation models (CUF, SLEUTH,

SimLand, UrbanSim…) could simulate the results of different land development policies, thus help policy makers to decide the best policy to regulate land development.

2.2.3 Big data and related techniques in urban planning

Big data can greatly enrich authorities’ understanding of how each individual stakeholders behave and their interest appeal. For example, the bus Smart Card Data in most cities records the commuting behavior of each individual passengers and the information of how each bus company operates. If urban planners make use of these data into their analysis and policy-making, they could have a better understanding of the current situation and major challenges these stakeholders are facing, and thus better balance their interests in their planning process. Therefore, as big data provides more information on each stakeholder’s behavior and demand in the city, the utility of such

17 data in planning process enables planners to take into consideration their interest, thus would benefit the negotiation among these stakeholders.

In addition, the wide-spread of big data is in some ways a decentralization of data collecting and analysis. This will stimulate some public interests of participation in negotiation in the planning process. And the use of big data in urban planning will move a step forward to the decentralization of authority, the weakening of political power/influence and thus contribute to the ‘bottom-up’ approach for planning.

Besides big data itself, the techniques related to big data also play a critical role in the negotiations in the urban planning process.

First of all, some techniques provide the channel for different stakeholders to negotiation, and enables group collaboration (sometimes from long distance) in the urban planning process. One of the most representative techniques is the Participatory Planning

Geographic Information Systems (PPGIS). PPGIS serve data with spatial reference to a wide audience via the Internet (Hancl, 2007). The data presented in this way include basic data for analysis such as property data, demographic data, cultural and natural heritage, and also planning scenarios such as master plan and investment areas location.

With access to these data, each stakeholder can conduct their own analysis considering their own interest using the tool provided, understand the current situation and planning scenarios, comment on different scenarios of city development, and participate in the urban planning process.

In addition, some techniques can help planners to quickly process the large data volume and generate the visualization of their planning scenario based on data analysis.

The examples are the Community Viz and What if (Brail and Klosterman, 2001). Both applications allow for simulation of how a place would change in the near future after introducing parameters describing current state and planning conditions (Hanzl, 2007).

The software would quickly generate the possible results of each land use alternatives and help the public (who does not have much professional training) to understand the potential environmental, economic and social impacts of different planning scenarios. A real-world case is a Paint the Town system that has been applied in 77 communities and

271 suburbs of Chicago. During meetings, officials and citizens negotiated over propositions of use of land and the character of development, and the tool allowed for prognosis presentation, the creation of scenarios and gathering information on local expectations, thus support the discussion and collaboration of every stakeholder involved.

Last but not least, big data and its relevant techniques enable the establishment and implementation of urban models that consider different stakeholders. For example, some multi-agent models have been used to provide technical support for urban planners to understand the current situation in the city they are going to plan. In these models, different stakeholders are represented as ‘agents’, and their demands are introduced into the model as parameters. The models are then able to model the joint behaviors of different agents, and simulate the results considering the behaviors of every agent.

Therefore, these models could provide technical support and suggestions to the negotiation of different stakeholders.

2.2.4 Summary of data analytics in urban planning

In summary, the new data environment and technology advancement recently has triggered interest in quantitative approaches and has equipped this powerful approach with large-volume find-scale data, advanced technical support and broader application potential. Analytic approaches can help facilitate urban planning in the following aspects:

(1) Determine expected urban growth rate: urban analytics could help urban planners figure out the expected population size in the near future, and thus to decide the amount of urban land development to meet the demand of population increase. In addition, some land-use models, such as the CUF (Landis, 2001) and What if?

(Klosterman, 1999), could analyze the land-use suitability in a certain area based on some pre-defined rules, thus help urban planners work out specific urban growth strategies to decides the amount of land to develop in the specific urban area.

(2) Figure out factors that influence urban growth: statistical methods and GIS analysis tools have been widely used to figure out how different factors would influence urban growth. For example, some regression models could give urban planners a quantitative measure as to what extent the population growth, transportation networks, geology conditions, or other factors could result in urban sprawl.

(3) Simulate urban growth in near future: another aspect of the use of analytic approaches is to simulate possible urban growth in the next several years. Cellular

Automata (CA) model is a representative tool. Using CA model, urban planners can generate the urban growth result regarding both social-economic conditions in the city and the policy/planning scenarios. For example, the Beijing City Lab in China has been

20 using CA model to work out the urban growth direction and possible layout of urban growth boundary in the next several years, and thus provide suggestions for urban planners to control urban sprawl in Beijing (Long et al., 2008).

(4) Simulate possible results in urban growth is certain development policy is conducted: some quantitative models can quantify different development policies or planning scenarios put forward by urban planners, to simulate the possible results in urban growth if the policy is carried out. This could evaluate the urban growth policies before they are being implemented, to avoid potential side-effects of the policies. Also, the quantitative models could simulate urban growth results for different urban plan scenarios, thus to provide a scientific reference for authorities to compare and select the best scenario (Long & Shen, 2015).

(5) Monitor urban growth: quantitative approaches could also help urban planners and managers to monitor urban growth and manage the land development. GIS and remote sensing techniques could generate detailed maps of land use in every several days, and help urban planners monitor the urban growth timely. This is significant in the implementation urban growth policies and the management of urban land development.

However, existing quantitative urban studies have seldom been applied to real- world urban planning practices. There are mainly three reasons. Firstly, quantitative urban studies are more effective in short-term analysis and prediction, but often fail to simulate long-term change accurately, as the prediction of parameters under the fast- changing situation is very challenging. Secondly, previous quantitative studies mainly focused on evaluation and problem identification of certain urban phenomenon, without

21 drawing implications for urban planning. Thirdly, there is a huge gap between traditional urban planning practice (including planning system, planning process, the cognition and skills of urban planners, etc.) and quantitative methods, which also hampers the adoption of quantitative tools in urban planning. Therefore, the goal of this research is to seek approaches that enable the application of quantitative urban studies to urban planning/design.

2.3 Positive and Normative Dimensions of City Science

As articulated by Batty (2013), we are now entering the era of a new city science.

Compared to the old city science that based on a more static and cross-sectional view of cities, the new science of cities studies urban issues based on the complexity theory with a more disaggregated, evolving and bottom-up thinking (Batty, 2013). The new city science pays more attention to dynamics of change and the interactions in cities.

In his recent work The New Science of Cities, Batty (2013) discussed the positive

(focusing on ‘what is’) dimension of the city science and the normative (focusing on

‘what should be’) aspect of it. In the field of urban studies, the positive approach, or empirical approach, could be related to urban modeling and spatial data analysis, while the normative approach, or idealistic approach, is more in line with the tradition of urban design. In the new science of cities, these two dimensions of city sciences are not mutually exclusive, and their tools and methods could be effectively combined to examine urban issues. With this combination, the positive tools could provide empirical support for normative goals, while the normative design targets would act as a guidance

22 for positive urban studies. Therefore, it is possible and could be powerful, to combine quantitative urban studies with urban design to achieve the quantitative urban design.

Inspired by this idea, this research is contextualized under the overarching goal of linking the positive (i.e. quantitative data analysis and urban modeling) with the normative (i.e. urban design targets) dimensions of the emerging science for cities, and more specifically, to integrate the ‘what if’ and ‘what should be’ issues in urban studies.

2.4 Theoretical Framework Linking Spatial Data Analysis with Urban Planning

The separation of quantitative studies and urban design in Chinese cities is now causing more and more problems in the recent trend of fast urbanization. Without applying the quantitative geographic analysis to provide evidential support, many urban designs are simply based on the experiences of urban planners or preferences of a government leader. Also, the traditional physical design always takes relatively longer time to be processed, so is unable to react immediately to the fast urbanization nowadays.

It is very common in Chinese cities that when a design is finally put forward, the situation has already changed. With regard to the quantitative approaches themselves, the lack of application to the urban design has limited their further development. For example, the Cellular Automata (CA) model, a very useful tool to simulate urban growth, has seldom been applied to real-world planning and is nearly ‘outdated’ in academia today.

Referring to the positive and normative dimensions of the new science of cities, this research proposes a framework that links ‘positive’ spatial data analysis with the

‘normative’ urban design. Compared to the ‘old’ city science, the new science of cities study cities not only as space and place, but also as systems of networks and flows.

Therefore, this research will be conducted from the perspective of both space/place and networks/flows in different Chinese cities, and will invoke the application of urban modeling and spatial analysis in the field of urban design on three levels concerning the city as a whole (space), community (place) and transportation systems (network) (Figure

3).

Figure 3. Theoretical Framework: Integrating the Two Dimensions of City Science

In the conceptual framework of this study, the positive dimension of new city science refers to empirical urban studies, while the normative dimension is urban design targets. The effort to integrate the two dimensions of city science will be conducted from three perspectives: space, place, and network. From the space perspective, cellular

24 automata model will be applied to evaluate the land-use performance of existing design scenarios empirically, and a new design scenario targeting at controlling urban sprawl will be generated based on the empirical studies. At the place level, positive urban study methods such as questionnaire survey and statistical analysis will be carried out to provide empirical support for the mixed-use development of communities, and the development processes that achieve the design goal of mixed-use will be compared and analyzed from a normative sense. As for the network aspect, the normative design of bus network that promotes accessibility will be achieved based on the empirical studies of bus smart card data and network accessibility measures. The three case studies conducted from the three perspectives will all have a positive process and normative design target, and thus will be a trial to integrate the positive and normative urban studies in the broader context of the emerging urban science.

Chapter 3: Space Perspective - Integrating CA Modeling with Geodesign

From the space perspective, the goal of this chapter is to generate the design scenario of urban growth to meet the need of fast urbanization. To achieve this goal, a geodesign-based Cellular Automata (CA) model is put forward. CA model is a powerful simulation tool to study complex urban systems. Although there has been a considerable amount of CA-based modeling work reported in the literature, few have been used as actual planning or policy-making tools. Instead of merely applying CA models to positive urban studies, which typically requires more accurate predictions, this chapter advocates the integration of positive (i.e. urban simulation) and normative (i.e. design) city science via the application of CA modeling to geodesign. By combining CA models with the framework of geodesign and targeting the positive urban simulation at urban design, we can address the problem of “unapplicable CA models”. On the other hand, CA modeling provides the empirical support to evaluate different design scenarios and generate new design.

In this chapter, a geodesign-based CA model is developed and six urban design scenarios are integrated into the model, to test the urban growth impacts of these scenarios and generate a new design scenario accordingly. Using a case study of urban growth in Changping District, Beijing, I demonstrate that the normative and the positive dimension of Batty’s new science for cities can be integrated. Geodesign, as a new 26 conceptual framework, is helpful for streamlining the CA modeling process to evaluate different design scenarios and design new scenario. In return, CA models could also contribute to the geodesign process by offering a tool for simulating and evaluating impacts of different design scenarios on urban growth.

3.1 Research Context

With the development of GIS, complex systems modeling, and the explosive growth of big data, recent years have witnessed renewed interests in simulating urban growth by urban modeling. Indeed, even a new science of cities is emerging as a result of these new developments (Batty, 2013). Urban models have developed from structural models to static models to dynamic models. As a dynamic model derived from biological theories and closely combined with GIS, Cellular Automata (CA) models are widely applied to individual decision-making and large-scale urban change simulation (Batty and

Xie, 1994; White and Engelen, 1997; Clark and Gaydos, 1998; Straatman et al., 2000;

Torrens and O'Sullivan, 2001; Yeh and Li, 2002). The concept of cellular automaton was originally discovered in the 1940s by Stanislaw Ulam and John von Neumann, but it was not until the 1970s that the interest in this subject expanded beyond academia. In the

1970s, Conway put forward “The Game of Life”, which contributed a new method to modeling on an infinite grid of cells. In “The Game of Life”, the “life” of each cell is subject to a series of local rules pertaining to the birth of new cells and the death of existing ones (Gardner, 1970). From such simulations, there emerged a variety of different models. This branch of models is called cellular automata models (Batty and

Xie, 1994). In general, CA models are composed of four principal elements: a lattice, a set of defined status, neighborhoods defined by the lattice, and transition rules. The status of a certain cell will be influenced by its neighborhood. The invention of computers has triggered the development of CA models, and these models are applied to the simulation of physical reality.

CA models have been used to study urban systems at nearly the same time with the creation of “The Game of Life”, when Tobler (1970) formalized and carried out an informal cell-space simulation of urban development and later defined it as cellular geography (Tobler, 1979). CA models for urban growth simulation should take into account not only the influence of neighboring cells, but also factors that depict the complexity of urban system development (Long et al., 2008). The CA models in urban simulation consider cities as organisms (Sui, 2010), and represent urban growth on a two- dimensional lattice of cells, where status of cells are depicted by land-use types and embodying processes of change in the cellular state over time are determined by the status and attributes of the cell itself as well as the neighboring cells (Batty, 2009). The infinite complicated driving factors of urban development are depicted by the constrained conditions, variables and transition rules of the models.

Recently, with the interests in space-time dynamics and the development of GIS,

CA models have sparked a flurry of applications in urban studies, from traffic simulation and regional-scale urbanization to land-use dynamics, historical urbanization and urban development (Torrens and O'Sullivan, 2001). The advantages of CA for urban simulation include its particular adaptability at dealing with spatial phenomena, highly decentralized 28 approach, affinity with GIS and remote sensing, the connection of form with function and pattern with process, the relatively dynamic process, the infusion of complexity theory, and its simplicity and visualization (Torrens, 2000). A number of researchers have applied CA models to simulate urban growth and predict future urban development.

White and Engelen (1997) applied CA models to urban planning and land-use patterns in

Cincinnati, Ohio, United States. Clark and Gaydos (1998) developed the SLEUTH model to simulate long-term urban growth in the San Francisco Bay and the Washington-

Baltimore areas. Batty and his collaborators conducted several studies using CA to study urban formation and expansion (Batty and Xie, 1994; Batty, 1997; Batty et al., 1997;

Batty, 2007). Wu and Webster (1998) used the multi-criteria evaluation (MCE) method to find the status transition rules in CA and applied the model for urban expansion simulations of Guangzhou, China (Wu, 2002). Li and Yeh used various methods to identify CA transition rules, with a constrained CA model to simulate sustainable urban growth in the Pearl River Delta and analyzed the uncertainty of CA (Li and Yeh, 2000;

Yeh and Li, 2001; Li and Yeh, 2002; Yeh and Li, 2002). Long et al. (2006; 2009; 2012) generated the Beijing Urban Development Model (BUDEM) using CA approach to support the urban planning and policy evaluation in the Beijing metropolitan area.

Despite the large number of empirical studies, few CA-based models have been used to test urban development policies (Batty, 2009) in a realistic way. Most CA models are an “experimented model” and are more appropriate as “metaphors” of urban growth rather than as actual simulators (Couclelis, 1985). As a result, CA models have rarely been adopted by the actual urban planning and policy-making processes.

Therefore, it is essential to find a new framework for CA model to facilitate its broader applications.

From the space perspective, the positive dimension of the new science of cities, or empirical approach, could be related to urban modeling and simulation, while the normative dimension, or idealistic approach, is more in line with the tradition of urban design. According to Batty (2013), these two dimensions of city sciences are not mutually exclusive, and their tools and methods could be effectively combined to examine urban issues. Therefore, it is possible, and could be powerful, to combine CA models with geodesign, which is a representative of normativism and studies the world

“as it should be”.

In this chapter, through the case study of urban growth geodesign in Changping, I demonstrate that the CA model application framed by geodesign could work well in designing process and the evaluation of different designing scenarios, and whereas, CA models could in some ways contribute to the geodesign process by offering a physical simulation tool. The CA simulation process framed by geodesign would build the bridge between the positive and normative dimensions of the city science.

The rest of this chapter is organized as follows. After the brief introduction, section 3.2 raises the conceptual framework of this chapter by linking the positive and normative dimensions of the new science for cities through a geodesign-based CA model; in section 3.3, the data and methods used in the case study of Changping, Beijing are introduced; section 3.4 represents the results, with section 3.4.1 analyses the simulation results and evaluations from the geodesign-based CA model, which is the positive 30 dimension of the city science, and section 3.4.2 focuses on the normative dimension via the generation of new design scenario from previous positive analysis; finally, section 3.5 summarizes the conclusions and puts forward an iterative framework of geodesign-based

CA modeling for urban development simulation.

3.2 Geodesign-based CA Modeling Framework

3.2.1 Geodesign

Geodesign is the thought process comprising the creation of entities in our geo- scape, or more simply, the design in geographic space (Miller, 2012). It brings geographic analysis into design process (Dangermond, 2009), provides a design framework and supporting technology to combine geography with design, and is considered as a connection between planning and geographic studies. Geodesign is a new concept especially after the emergence of Web 2.0, but geodesign activities have been with us “since the beginning of time” (Miller, 2012). Steinitz (2012) developed a comprehensive framework for practicing geodesign as applied to regional landscape studies. This framework consists of six questions asked (explicitly or implicitly) in three iterations during any geodesign practices, which are concerned with the description, operation, evaluation, change, effectiveness and decision of the study area (Figure 4). The answers to these questions are six models. The first three models comprise the assessment process, which are representation, process, and evaluation models, looking at existing conditions within a geographic context. The other three models constitute the intervention 31 process, including change, impact and decision models, focusing on how the context might be changed, the potential consequences of those changes, and whether the context should be changed. This framework of six questions and models is our major concern in this study to guide the CA simulation process, and is applied to connect the positive and normative dimensions of city science.

Figure 4. The Framework of Geodesign [source: Steinitz (2012)]

There are three reasons that we introduce geodesign framework to rejuvenate CA models. Firstly, the six questions put forward by Steinitz (2012) include the procedure of

32 scoping, analyzing and designing, and go through the steps of obtaining and organizing the data, calibrating and testing the process models, evaluating past and present conditions, proposing and simulating future changes, assessing and comparing the impacts of each change via the process models, and finally, making a decision. This procedure could be accomplished via CA simulation. Secondly, the main idea of geodesign framework is to process models under change conditions and enable the design of alternative futures (Dangermond, 2009). Therefore, if we are able to embed different change conditions into CA model, and run the model to predict the outcomes of these different conditions, CA models would act as a powerful process model to assist geodesign. Last but not least, one problem of CA models is the incapability in model validation, and the models always fail to accurately simulate the complicated urban expansion and land-use change process. However, CA models would work well in geodesign by focusing on generating a new design from alternative development scenarios, which requires more creativity/imagination but less precision/accuracy.

3.2.2 Geodesign-based CA modeling

The framework of geodesign is invoked here as a conceptual framework to guide the CA modeling process and thus to build the bridge between the two dimensions of city science (Figure 5).

Figure 5. Conceptual framework: integrating CA modeling with geodesign

Geodesign studies the world as “what it could be”, and tests how possible alternatives will change the results. Urban CA model studies the world as “what it is”, and the key is to examine the impacts of driving factors on urban development and create transition rules for future prediction of different design scenarios. In this framework, the six questions/models of geodesign are applied to guide the CA modeling process with the goal of designing, including the representation of study context via examining the influence of driving factors on urban development (representation models), processing

34 how the urban development of study area operate using CA simulation (process models), evaluating whether the current context is working well based on several critical indicators of land-use impacts (evaluation models), the generation of new design scenario according to evaluation and running the CA model to produce the results (change models), evaluating the performance of the new design and compared with previous scenarios to reveal the differences caused by new alternatives (impact models), and finally work out the final decisions of design (decision models). This procedure applies CA modeling through the six questions asked in geodesign framework, and imbeds both the positive

(representation, process and evaluation of the existing design scenarios) and normative

(alternative design generation, impact evaluation and decision-making) city science.

3.3 Data and Methodology

3.3.1 Study area and design target: urban sprawl control

Changping District, an important district and new development zone in Beijing, is chosen as the study area to test the application of geodesign-based CA models (Figure 6).

As the capital of China, Beijing has been through very fast urbanization in the past 30 years, and is still in urgent need of urban construction and development. In the Beijing

Master Planning 2004-2020 drafted by Beijing Municipal Commission of Urban Planning

(2005), Changping District is assigned the responsibility for the sustainable development and healthy urbanization of the whole Beijing city, with the mission to undertake the residential and industrial functions used to be performed by the central city. Driven by

35 this planning policy, Changping district has been through very fast urban growth (Figure

7), and this trend will go on in the next decade. At the same time, located at the northwestern of Beijing, the large mountainous areas in Changping are considered the second green isolated belt that preserves the ecology and environment of the whole

Beijing city, and the rivers are serving as the upstream of the water source in Beijing. The design of Changping needs to take into consideration not only economic efficiency, but ecological preservation and environmental influence. Therefore, the design target of this study is to control urban sprawl and minimize the negative impacts of urban growth.

Figure 6. Study area: Changping district in Beijing, China 36

Figure 7. Urban growth of Changping district from 2005 to 2012

In the design of future urban development in Changping, six scenarios are most representative (Figure 8). Scenario A is a City Master Planning approved by the State

Council of PRC in 2005. Scenario B is a controlled development scenario, which encourages the prevention of excessive sprawl in central Beijing and the promotion of new towns development via spatial control (eg. land policy) and political zoning.

Scenario C, the sprawling scenario, on the other hand, promotes developments in the central Beijing while the development in new towns is controlled. This scenario is similar to the “pancake development” in Beijing in recent years. Scenario D is a grape-clustering development design generated by Beijing City Lab (BCL). This design scenario encourages the urban development along traffic lines or around small towns. Scenario E is an eco-preserving design that focuses on ecological preservation, and tries to avoid the encroachment of urban construction into ecological spaces, resulting in a more dispersed urban form. Scenario F is a design scenario constrained by topographic features.

Figure 8. Six design scenarios of the study area

3.3.2 Data

The urban development in Changping District is influenced by social as well as accessibility factors. The factors stated in Table 1 will be considered and put into the geodesign-based CA model for future developing prediction.

Table 1. Driving Factors of Urban Development in Changping District

Types of Driving Data Variables Data Type Data Factors Factors Source Changping Numeric statistical Population Population (standardized) yearbooks 2006-2013

Social Factors Harvested Housing Numeric Housing from Ganji Prices (standardized) Website

Derived Distance to Numeric from GIS D_city new cities (standardized) vector layers

Derived

Distance to Numeric from GIS Accessibility D_town main towns (standardized) vector Factors layers

Derived Distance to Numeric from GIS D_roads roads (standardized) vector layers

Derived Distance to Numeric from GIS D_rivers rivers (standardized) vector layers

Derived Distance to Numeric from GIS D_center city center (standardized) vector layers

Status of 8 Binary (1: Derived Neighboring nearest neighbor urban built-up; from GIS Factors neighbors 0: non-urban) raster layers

Changping Basic Numeric (total Constraints Farmland government farmland area) documents

Urban growth rate=elastic coefficient of Changping Estimated Rate Numeric urban growth government Population (1.12) *estimated documents population growth rate Steep Binary (1: Changping Mountains plains; 0: steep Restrictions Restrict government and water mountains and documents body surface water)

master Binary (1: Changping planning scenario_a urban built-up; government scenario 0: non-urban) documents

controlled Binary (1: Changping Designing development scenario_b urban built-up; government Scenarios scenario 0: non-urban) documents

sprawling Binary (1: Changping development scenario_c urban built-up; government scenario 0: non-urban) documents

grape- Binary (1: clustering Beijing City scenario_d urban built-up; development Lab 0: non-urban) scenario

eco- Binary (1: Changping preserving scenario_e urban built-up; government scenario 0: non-urban) documents

terrain- Binary (1: Changping limited scenario_f urban built-up; government scenario 0: non-urban) documents

Binary (1: Changping Dependent Urban lulc urban built-up; government Variable Development 0: non-urban) documents

The data used come from four sources: GIS dataset, statistical data, governmental documents and website harvesting. All data was converted into a 30m*30m raster format, using the same projection system and spatial reference.

(1) GIS Dataset: The base dataset for analyzing and modeling is the 30m*30m

raster data frame of Changping. The accessibility factors, including the spatial

distance to new cities, towns, roads, rivers and Tian’anmen Square (considered

as the center of Beijing city), are derived from GIS vector layers. All the

distance values have been standardized.

(2) Statistical Data: county-level Changping Demographic Census Data for the

population is obtained from Changping statistical yearbooks 2006-2013. All

population data are converted to raster dataset and standardized.

(3) Government documents: Scenarios A, B, C, E, F, areas of basic farmland, maps

of steep mountains and the land-use maps are obtained from Changping

Planning Department. Scenario D is the raster dataset generated by Beijing City

Lab. The estimated population is provided by Changping New Town Planning

for 2020 (Beijing Planning Committee, 2007), which is set to 1074000 in 2020.

We calculated the estimated population growth rate and multiplied the elastic

coefficient (1.12) to obtain the constraint of urban growth rate.

(4) Website Harvesting: housing price data is harvested from Ganji Website

(Beijing City Lab, 2013), which is an important website for real estate

transaction in China. 538 points of housing location as of Oct. 2013 are

obtained, together with their housing price and established date. The data of

housing price is interpolated and converted to raster dataset using Kriging

method and standardized.

3.3.3 Methods

In a standard urban simulation process using CA models, the state is considered the main attribute to describe the development of a cell, and the most general state for a cell is developed (urban) or undeveloped (rural). If we use St {i, j} to represent the state of the cell (i, j) at time t, then a standard CA model may be generalized as follows:

S푡+1{푖, 푗} = 푓 (S푡{푖, 푗}, 푁),

1, developed/urban, S푡{푖, 푗} = { 0, undeveloped/rural,

where S is a set of status of the cellular automata; N is a neighborhood of cell (i, j) providing input values for the function 푓; and 푓 is a function that defines the change of the state from t to t+1. In our study, neighborhood (N) is defined as the states of cells from the Moore neighborhood (3*3 neighborhood window around the cell under concern). Heterogeneous neighbors are used to improve the actuality and rationality of the model, and the neighborhood influence is incorporated as follows:

∑ 푤{푖, 푗} ∗ 푆푡{푖, 푗} Ω푡{i, j} = 3∗3 , ∑3∗3 푤{푖, 푗}

where w{i, j} is influenced by transportation systems, which means that the cells connected by traffic lines will have a more significant influence on each other.

More sophisticated CA systems have further assumed a relation between cell states (developed or not), development probability and development suitability (White and Engelen, 1997; Wu and Webster, 1998):

S푡+1{푖, 푗} = 푓(푃푡{푖, 푗} ),

P푡{푖, 푗} = 푓(퐷푆푡{푖, 푗} ),

where S {i, j} is the development state at location (i, j); Ps {i,j} is the probability of transition to the state S at the location; and DSs {i, j} is the development suitability that describes the potential of a cell to be developed as urban land use; 푓 is a transition function. Therefore, the standard CA model may be further expressed as:

S푡+1{푖, 푗} = 푓(S푡{푖, 푗}, 퐷푆푡{푖, 푗} , Ω푡{i, j}).

To integrate the two dimensions of city science, we need to incorporate the normative dimensions into the model. This new paradigm for thinking about city science is accomplished through the implementation of urban planning or policies regarding urban growth. It is also possible to embed some constraints and restrictions in the transition rules of CA so that urban growth can be rationalized according to some pre- defined design criteria, thus the final model becomes:

S푡+1{푖, 푗} = 푓(S푡{푖, 푗}, 퐷푆푡{푖, 푗} , Ω푡{i, j}, P, 퐶푂푁푆푡{푖, 푗}) ∗ 푅퐸푆푇푡{푖, 푗},

where P is the planning or design scenarios; 퐶푂푁푆푡{푖, 푗} is the constraints that have some influences but do not have critical restrictions on the urban development;

RESTt {i, j} is the restrictions that have crucial effects on the modelling process. For example, the estimation of population growth, the policy of total farmland to be preserved and a possible threshold of urban development may constrain the urban growth rate, but steep mountains and water bodies would be restrictive constraints that restrict urban development from taking place.

MCE (Multiple Criteria Evaluation) techniques can be used to estimate the development probability of each cell. The techniques can also be applied to determine constraints, restrictions and development suitability that are dependent on several social- economic and accessibility factors:

∑푙 푡{ } 푡 푘=1 푊푘퐷푆푘 푖,푗 퐷푆 {푖, 푗} = 푙 ), ∑푘=1 푊푘

∑푚 푡 { } 푡 푘=1 푊푘CONS푘 푖,푗 퐶푂푁푆 {푖, 푗} = 푚 , ∑푘=1 푊푘 44

푡 푛 푡 푅퐸푆푇 {푖, 푗} = ∏푘=1 푅퐸푆푇푘 {푖,푗}.

Various methods have been developed to obtain the weights of spatial variables in the MCE, such as CUF and CUF-2 (Landis and Zhang, 1998), analytic hierarchy process

(AHP) method (Wu and Webster, 1998), logistic regression (Wu, 2002), nested loops

(Clark and Gaydos, 1998) and monoloop (Long et al., 2009). In this study, we integrated the methods of Wu (2002), Clark and Gaydos (1998) and Long et al. (2009), and used the bis-loop method to determine the weights in MCE. All the spatial variables except neighbor and threshold of development were included in the logistic regression equation, and the corresponding coefficients (weights in the MCE) were obtained. Then, the weight for neighbor factor (wn) and the threshold of the cell to develop as urban (thed) were calculated using the bis-loop method. In total, 800 possible combinations of these two variables were accessed while keeping the other coefficients constant. For each combination, we simulated the urban growth from 2005 to 2012, and by comparing with the actual land use data in 2012, goodness-of-fit (GOF) for every combination was calculated to figure out the best combination. To minimize the looping time, we used the parallel tasks, and the total looping time is around 200 minutes.

After obtaining all coefficients and variables, CA model was run to predict the possible urban form of the year 2020 under the six design scenarios. Based on the predictions, we evaluated the performance of six design scenarios. Based upon the land resource impact indicators of urban sprawl proposed by Hasse and Lathrop (2003), we examined four critical land use impacts associated to urban development: (1) urban sprawl rate; (2) loss of prime farmland; (3) loss of core ecological habitat; and (4) 45 geological conditions of developed areas. The Land-use Impact Indices (LIIs) of six scenarios were then calculated from the four critical impacts for each of Changping’s 321 administrative villages. With the design target that minimizes the negative impacts of urban development, we then determined the scenario that generates best LIIs in each village, and integrated them to generate a new geodesign scenario.

3.4 Results

3.4.1 Positive dimension: simulation and evaluation

Figure 9 shows the GOFs obtained from bis-loop. By comparing the GOF for each variable combination, wn and thed of different scenarios are determined. The highest

GOF of all the combinations is around 82%, which means around 82% of the cells could be correctly simulated by our model.

Figure 9. Goodness-of-fit (GOF) of all experiments.

After obtaining all the coefficients in the model, the geodesign-based CA model was processed to simulate urban growth in 2020. According to the results, Changping will experience fast urban growth in the next few years, and most of the southern and eastern part will become urban areas. Urbanization will occur significantly in the eastern

47 and southern direction, while the major part of western and northern mountainous areas will remain as rural areas.

With regard to urban growth under six scenarios (Table 2), we can conclude that urban growth rate under scenario A (master planning) and D (grape-clustering scenario) is lower (around 16% compared to 2012), and only the central Changping and the southeastern part will have significant urban sprawl. Scenario F (terrain-limited scenario) will lead to most severe urban sprawl, with the urban growth rate as 21.83%. Scenario B

(controlled scenario) and C (sprawl scenario) will also suffer from fast urban growth.

Therefore, with the target of controlling urban sprawl, simply create a controlled- development scenario through spatial zoning may not be sufficient. A specific design of certain urban form (eg. grape-shape clustering form) may achieve a better urban performance.

Scenario Scenario Scenario Scenario Scenario Scenario A B C D E F Urban Area in 2020 (km2) 448.20 460.31 458.67 446.27 452.38 467.60 Urban Area in 2012 (km2) 383.82 383.82 383.82 383.82 383.82 383.82 Urban Growth 16.77% 19.93% 19.50% 16.27% 17.86% 21.83% Table 2. Prediction results of urban development in 2020

To evaluate the six scenarios and thus to provide some insights to the new geodesign, we examined four critical land use impacts of the predicted urban development in 2020 under six scenarios. The four land-use impact indicators are: (1) 48

Sprawl Indicator (SI) that depicts urban sprawl based on the density of urban areas; (2)

Farmland Indicator (FI) that calculates the percentage of basic farmland being occupied by urban development; (3) Ecological Indicator (EI) that evaluates the performance of design scenarios on ecological preservation with regard to the green space being occupied by urban development; (4) Geological Indicator (GI) that measures the geological conditions of the urban areas in 2020. Based on the four indicators, we then calculated the general Land-use Impact Index (LII) to evaluate the outcomes of six scenarios. The smaller the indicators, the better performance that scenario has.

Table 3 summarized the results of our evaluation based on land-use impact indicators. As we can see, scenario A (master planning) and D (grape-clustering scenario) most significantly control urban sprawl; urban development under scenario A and C occupied less farmland in 2020; scenario E (eco-preserving scenario) performs best in ecological preservation, and its geological condition of urban areas is best. To our surprise, according to our land-use impact indicators, scenario B that is targeting sprawl control and scenario F that assigns the least urban areas due to terrain limitation have the worst outcomes in urban sprawl control, farmland preservation, ecological protection and geological conditions. This conclusion further demonstrates our previous statement that a design scenario that simply controls urban sprawl through political zoning or spatial constraints may not achieve the target as desired.

Scenario Scenario Scenario Scenario Scenario Scenario A B C D E F Urban Area in 2020 (km2) 448.20 460.31 458.67 446.27 452.38 467.60 SI 0.3387 0.3478 0.3466 0.3372 0.3418 0.3533 Occupied Farmland (km2) 55.44 59.95 55.78 57.45 59.08 59.91 Total Farmland (km2) 98.58 98.58 98.58 98.58 98.58 98.58 FI 0.5624 0.6082 0.5658 0.5828 0.5994 0.6077 Occupied Greenspace (km2) 36.22 40.21 40.24 38.76 35.14 46.59 Total Greenspace (km2) 651.45 651.45 651.45 651.45 651.45 651.45 EI 0.0556 0.0617 0.0618 0.0595 0.0539 0.0715 Good (km2) 75.36 77.43 76.76 75.67 76.30 76.63 Not Bad (km2) 20.43 22.86 23.04 21.41 19.53 29.07 Bad (km2) 13.93 14.32 13.97 14.08 14.17 14.20 Worst (km2) 15.41 16.01 15.53 15.48 15.75 16.09 Total Urban Area (km2) 448.20 460.31 458.67 446.27 452.38 467.60 GI 0.1225 0.1250 0.1237 0.1247 0.1221 0.1292 LII 1.0792 1.1427 1.0978 1.1043 1.1172 1.1618 Table 3. Land-use Impact Indicators of six scenarios (smaller number indicates better performance)

We also calculated the SI, FI, EI, GI of every administrative village under six scenarios, and generated LIIs on village level accordingly (Figure 10). Based on the LIIs, we are then able to determine the weights of different scenarios on village level and create a new design scenario.

Figure 10. Land-use impact indices (LIIs) of the six scenarios at the village level

3.4.2 Normative dimension: a geodesign scenario

In city science, the normative dimension is related to design that focuses on “what should be” issues. Based on the previous positive/empirical approach of predictions and evaluations, we then generate a new geodesign scenario (Figure 11). The new scenario provides a clustering urban form, with four clusters of urban areas: central Changping, western mountain-front area, southern cluster and eastern cluster. Several small clusters are distributed around the four large urban agglomeration centers. This scenario has also put an effort to avoid the areas with the poor geological condition and the destruction of ecological habitats.

Figure 11. A new geodesign scenario for Changping 52

Figure 12 shows the prediction of urban development in 2020 under this new geodesign. The urban sprawl still exists, as an unavoidable trend, but under better control.

Also, the examination of land-use impact indicators (Table 4) demonstrates that our new design has better performance in all the four critical aspects of urban development impacts and achieves a lower LII.

Figure 12. Simulation of urban growth in 2020 under the new design scenario

Scenario A Scenario B Scenario C Scenario D Scenario E Scenario F Design SI 0.3387 0.3478 0.3466 0.3372 0.3418 0.3533 0.3320 FI 0.5624 0.6082 0.5658 0.5828 0.5994 0.6077 0.5704 EI 0.0556 0.0617 0.0618 0.0595 0.0539 0.0715 0.0555 GI 0.1225 0.1250 0.1237 0.1247 0.1221 0.1292 0.1210 LII 1.0792 1.1427 1.0978 1.1043 1.1172 1.1618 1.0788 Table 4. LIIs of new design and six previous design scenarios

3.5 Discussions and Conclusions

In this chapter, our primary goal is to explore the integration of the normative with the positive dimension of the new science of cities by framing cellular automata model under geodesign framework. We conducted a case study of urban growth simulation and design in Changping district, Beijing, and integrated six design scenarios into a geodesign-based CA model to simulate and evaluate the urban growth results of these design scenarios.

In the above sections, it is demonstrated that geodesign and CA models can be combined to integrate the positive and normative dimensions of city science. In this section, we further discussed the geodesign-based framework of CA modeling through an iterative process.

Geodesign is an iterative process that occurs rapidly (Miller, 2012), and is supposed to be an iterative framework. In Steinitz’s (2012) framework, the idea of iteration has been represented by going through the six steps both forward and backward

54 for at least three times. However, a more detailed framework is necessary to depict the iterative process of geodesign-based CA modeling.

Therefore, the iterative framework of our geodesign-based CA modeling includes three components: scoping, analyzing and designing (Figure 13). In Figure 13, Arrows

①, ②, and ③ form the iterative cycle between scoping, analyzing, and designing; arrows ② and ④ describe the iterative forward and backward between designing and

CA simulation to generate the best design; arrow ⑤ shows the input of CA simulation

(existing design scenarios for analysis and evaluation); arrow ⑥ is the output (the new design scenario). Scoping is the identification of the study area and designing context, and the establishment of the decision-making criteria based on the knowledge about the study area. Analyzing, achieved by CA simulation, includes three main steps in CA modeling, that is, identifying driving factors of the change, obtaining weights of each factor by examining how the factors will change the results, and predicting the future based on the CA model simulation. These three steps in analyzing themselves form a sub- iteration, and take existing design scenarios as inputs for further analysis, evaluation and design. The third component, the designing process, focuses on evaluating existing scenarios and generating new scenarios as outputs. There are also three parts in designing process, including evaluation, generation of new design scenarios and decision implement. Again, these three steps in the designing process are iterated to work out the best design scenario. Scoping, analyzing and designing should be iterated to work out the final design, and there are also minor-iterations between CA simulation and design process if necessary.

Figure 13. The iterative framework of geodesign-based CA modeling for urban growth design.

Through the simulation of urban growth results of the six scenarios, we have discovered that urban areas will keep growing in Changping, and southern and eastern

Changping will continuously suffer from severe urban sprawl in the next several years.

Among the six scenarios, scenario A (master planning scenario), D (grape-clustering scenario) and E (eco-preserving scenario) behaves well in urban sprawl control, farmland and ecology preservation and geological conditions, while scenario B (development- controlled scenario) and F (terrain-limited scenario) will lead to severe urban sprawl and destroy basic farmland and ecology. This implies that a better design scenario should not only consider spatial constraints that are focusing on sprawl controlling, but also take into consideration ecological preservation and provide a good urban form. Based on the

56 prediction and evaluation of the six scenarios, we generated our own design scenario accordingly. The new scenario is verified to be beneficial for urban sprawl control, farmland and ecology preservation and also tend to locate urban development in areas with better geological conditions. This study implies a possible and powerful integration of positive (urban modeling) and normative (design) dimension of city science. The research reported here is a continuation of recent work related to geodesign in the context of the new urban science for designing a new sustainable future (Eikelboom and Janssen,

2015; Linde et al., 2015). We also believe further work is needed to link geodesign with critical GIS for urban applications as envisioned by Wilson (2015).

However, the accuracy of our model is affected by data insufficiency. As revealed by the statistical methods in Chinese Population Census, the statistic caliber of the population in China has been conducted frequently, resulting in the inconsistency of census data; and the transforming between different data formats also led to some unavoidable inaccuracy. But the results of our simulation is still interpretable and instructive even with the data whose quality is not sufficiently high, which implies that the framework of our model is applicable, and the integration of normative and positive dimension in city sciences is significant. Future work should mainly focus on including more institutional policies into the model and the evaluation of geodesign, as well as quantifying and testing the uncertainty of different geodesign scenarios in the simulation and evaluation process.

Chapter 4: Place Perspective - Integrating Questionnaire Analysis with Mixed-use Development

From the place perspective, the design goal is mixed-use development. Mixed-use development has been widely accepted as a strategy in urban planning to address the problems resulted from the traditional zoning in the West, and it has also been increasingly adopted in many Chinese cities in recent decades with quite different results.

However, few studies have provided empirical support to the claimed benefits of mixed- use development and little is known about the process to achieve the desired benefits of mixed-use. This chapter investigates into three typical urban development models commonly used in China – “top-down” centrally-controlled development model,

“bottom-up” individual-dominant development model, and “bottom-up” collective- dominant development model.

Using southern Changping of Beijing as a case study, where the mixed-use development has been adopted in the past decade, we conduct a systematic evaluation of the three approaches and assess the impacts of mixed-use on urban development in

Beijing. By conducting questionnaire analysis and one-way analysis of variance

(ANOVA) to compare the job-housing pattern, career development of residents, sense of community and community vitality of the three models, the paper discovers that the community under “bottom-up” collective-dominant development model effectively

58 achieved mixed-use development, while the “top-down” centrally-controlled development may lead to functional division and the “bottom-up” individual-dominant developed community ended up in disorder and chaos. Our findings indicate that under current policy framework and development trends in China, the “bottom-up” collective- dominant development model and social inclusion would be an effective way to achieving the intended goals of the mixed-use development.

4.1 Research Context

Mixed-use development encourages urban planners and developers to form the compact, walking-friendly and mixed communities by fusing together different functions such as commercial, residential, and recreational land uses, so as to improve the economic and social vitality at the community level (Lynch, 1984). While the term frequently appears in the planning and real estate literature, the definition of mixed-use development is an ambiguous, multi-faceted concept (Rowley, 1996), and is rarely elaborated upon with substantive and empirical support (Herndon, 2011). The definition developed by the Urban Land Institute (ULI) is consistently referenced in the literature

(Herndon, 2011), which defines a mixed-use project as a coherent plan with three or more functionally and physically integrated revenue-producing uses (ULI, 1987). We follow this definition of mixed-use in this chapter.

The concept of mixed-use development is proposed against the growing functional division in urban design and planning in Western cities in the 20th century

(Lynch, 1984). Influenced by the principles of functionalism, zoning had been firmly 59 entrenched since the 1920s in the European and North American cities as a strategy to increase efficiency and safety by separating incompatible land uses (Hoppenbrouwer &

Louw, 2005). Zoning had played an important role in the reconstruction and recovery efforts after World War I. However, like many other well-intended urban policies and planning initiatives, functional zoning created many of its own problems as it was repeated mechanically in these cities, such as congestion, pollution, urban sprawl, workplace-residence separation and the loss of urban vitality (Rowley, 1996; Grant,

2002; Qiu, 2009; Herndon, 2011). The rationale behind the functional division in urban planning was challenged when Jane Jacob (1961), an influential urban scholar and critic, published her classic, The Death and Life of Great American Cities, in which she argued that the mix of diverse uses created vibrant and successful neighborhoods. As urban renewal was proposed in the 1960s in the US and European cities, the ideas of rebuilding the mixed-functional urban space was gradually accepted by urban planners in western countries, and the mixed-use development became one of the key components of the modern urban theories since the 1980s (Burton, 2000; Handy, 2005; Chen et al, 2008;).

As the most representative ideas of mixed-use, the new urbanism put forward two modes to achieve mixed-use development (Rodríguez et al, 2006; Duany et al, 2010):

Traditional Neighborhood Development (TND) mode (Grant, 2006). With the emergence of sustainable development as a new development paradigm/strategy, urban sustainability has become widely adopted as a new practice, and the mixed-use development was widely considered as an important path towards the “compact city” and “smart growth”

(Barnett, 2007).

During the past three decades, Chinese cities have repeated similar mistakes as western cities. In the urban fringe areas where rapid urban sprawl took place, large-scale single functional areas quickly come into being, which brought about severe urban problems (Qiu, 2009; Zhuang & Ren, 2011). The villages around these urban fringe development zones, however, were transformed into the urban villages under the spontaneous construction of individual villagers, with overloaded population, insufficient infrastructure and disordered management (Wei & Yan, 2005; Zheng et al, 2009; Wang et al, 2009). These problems in Chinese cities could be attributed, to a large extent, to the functional division in city reconstruction which is similar to western cities, as well as the particular land-use mechanism and urban planning systems in China. Therefore, extensive attention is paid to the mixed-use development in Chinese cities, especially in urban fringe areas (Ying, 2009; Li, 2010; Zhuang & Ren, 2011).

Support for mixed-use development has increased in the literature and by interdisciplinary researchers around the world. For developed countries such as the

United States, Canada, Japan and European countries, in which the urbanization level is high, mixed-use development has become a key element in both modern urban theories and planning practice. Studies provided evidence for the benefits of mixed-use development, treating it as a tool to address multiple urban problems (Garreau, 2011) and to realize sustainability and smart growth goals (Burton et al., 1996; Frey, 1999;

Calthorpe & Fulton, 2001; Barnett, 2007). Some scholars tried to develop theoretical frameworks for mixed-use development by working on the definition (ULI, 2003), dimensions and scales (Hoppenbrouwer & Louw, 2005), characteristics (Witherspoon et

61 al., 1976; Grant, 2002), feasibility (Wheaton, 2001; Dixon & Marston, 2003) and obstacles (Rowley, 1998). Empirical studies concerning mixed-use mainly focus on the experience of some mixed-development examples (Grant, 2002) and the detailed policies, regulations and community design (Schwanke & Flynn, 1987) in the West.

Studies of mixed-use development in Chinese cities, however, are still in the beginning stages. Theoretical research focuses on introducing the theories and experience of mixed-use development from the developed countries (Wen, 2009; Zhuang & Ren,

2011), discussing the importance of mixed-use development for the diversity, vitality and sustainability of modern cities (Qiu, 2009), and analyzing the possibility of successful mixed-use development in inner-city areas, urban fringe areas and decaying areas (Xing,

2005); studies about practice are mainly concerned with the generalization, evolution, mechanism and policies of functional-mixed communities (Zhu et al., 2010), as well as the evaluation of the existing mixed-use development experience (Yin, 2007; Liu, 2008).

All these studies explored the concepts and policies for mixed-use development, but few of them have conducted empirical research about effectiveness and mechanism of mixed- use development in Chinese cities.

4.2 Alternative Paths to Mixed-use Development

4.2.1 Design target: mixed-use development

The design target of my research on the place aspect is the mixed-use development of communities. As a popular urban design strategy, mixed-use 62 development has been practiced in many cities around the world, but the outcomes varied considerably, contingent upon local conditions and circumstance (Rowley, 1996; Grant,

2002; Dixon & Marston, 2003; Li, 2010; Herndon, 2011; Zhuang & Ren, 2011). For example, mixed-use development has been stated as one of the key strategies in the development of urban fringe in the General City Plan of Beijing (2004-2020), and has been conducted in Changping, Yizhuang and many other newly-constructed towns, but the results show significant disparity. Apparently, mixed-use development is not just the mix of functions, but a result of the conflict and compromise between different interest groups (Saich, 2000), which is influenced by different development processes. Therefore, a key question is: what is the most effective path to achieve the goals of the mixed-use development? This chapter will try to answer this question based on a case study of

Southern Changping in Beijing, China, an urban fringe area where the mixed-use development has been applied in the last several years. Three mixed-functional communities with different urban development patterns and disparate results are investigated, with the goal of teasing out the underlying mechanisms that ensure the effectiveness of mixed-use development.

4.2.2 Urban form and urban performance

According to Lynch (1984), three types of theory aim to explain the city forms.

Planning theory, also known as decision theory among planners, is about the development process that asserts how complex public decisions about city development are or should be made. The second type, the functional theory, attempts to explain why

63 cities take the form they do and how that form functions. Normative theory is the third type, which deals with urban performance - the generalizable connections between human values and settlement form, or how to define a good city. Different urban forms will lead to different urban performance.

The compact urban form comes from mixed-use development is considered as a significant agenda to create and maintain the goals of new urbanism: vital, beautiful, just, environmentally benign human settlements (Talen, 2005). Firstly, mixed-use development is a strategy for arranging the physical space that is required for society to function (Herndon, 2011). Moreover, this mixed-use form revolves around the desire to alter the undesirable growth patterns characterized by the traditional zoning. Furthermore, it “forms part of a strategy for sustainable development as well as a theory of good urban form, with the objectives of economic vitality, social equity, and environmental quality”

(Grant, 2002).

However, the mixed-use urban form is not a panacea, and the effectiveness of mixed-use development could not be absolutely assured. As Coupland (1997) points out,

“while some of the advantages of mixed-use can be accepted as absolute, others may or may not be true in certain circumstances”. In the previous experiences, cities have different performance even with the similar urban form characterized by compact and mixed-use. This suggests a topic worthy of exploration: what causes the different performances in the cities and communities pursuing the similar urban form of mixed-use development? Obviously, the effective urban performance requires not only urban forms, but more importantly, the proper development process to realize the urban forms, 64 functions and performance. Urban development in China is now at a crossroads.

Decentralization of decision making, market-led urban development initiatives, increase in the number of players/stakeholders and conflicts of different interest groups have challenged fundamentally the practice of urban planning (Yeh & Wu, 1999). In China, several stakeholders will participate in the urban development process, including central or local governments, community committees, companies, institutions, and individuals.

Admittedly, there are a few other stakeholders showing up in some specific cases, but they are excluded from our study for simplicity. From the leading stakeholder in the development process, we can distinguish three representative development models that dominate the modern urban planning and development in China. The three models are

“top-down” centrally-controlled development model, “bottom-up” individual-dominant development model and “bottom-up” collective-dominant development model.

In the “top-down” centrally-controlled development, the municipal government acts as a predominant decision-maker, while individuals can only passively involve in the development process. This development model comes from municipal authorities, represented by capitalist forces, central control and rational power. As a result of the decentralization after economic reform in 1978, China’s local governments began to play a dominant role in urban development through land finance, while the central government only acts as policy guidance. Therefore, the centrally-controlled development in our case is mainly controlled by local governments, which, compared to the urbanization dominated by individuals or collective entities representing individual interests, is still considered as a “top-down” process. Although still dominant in urban development, the

“top-down” model is challenged by two “bottom-up” development models: individual- dominant development and collective-dominant development. The individual-dominant development is a completely discursive process, in which self-conscious individuals

(including individual persons, companies or institutions) assert their own principles and complete their own development without any coherent goals. As a result, all personal developments are valid in the individual-dominant development, but these disorganized developments never take into consideration the consistency to the city master planning or how to “making sense together” of all individual developments. This development model mainly occurs at the urban fringe area in Chinese cities. Analogously, the collective- dominant development is a commonly-seen “bottom-up” process in the urban fringe area, while there is a collective entity that dominates the development. The collective entity is an agent that organizes personal decisions and bridges between central master planning and individual interests. The typical example of such entity in China is the villagers’ committee, which is elected by village residents and has a defined sphere of autonomy to manage public affairs (Kelliher, 1997). The collective-dominant development is neither intensive nor dispersive development, but a result of collective and interactive communications.

All the three development models have been carried out in the study area of this chapter. By conducting a case study of these three models, this chapter tries to demonstrate that different development processes will lead to different urban performances, even though they have the similar urban form of mixed-use development

(Figure 14). Furthermore, we will figure out which of the three development models most

66 effectively achieve the performance of mixed-use development. We state that the key to achieving a successful mixed-use community should be the development process, other than simply build the compact city forms.

Figure 14. Development process, urban form, and urban performance

4.3 Study Area, Data, and Methodology

4.3.1 Study area

Mixed-use development occurs in different scales. For example, Hoppenbrouwer and Louw (2005) made distinctions between buildings, blocks, districts, and cities. In

Rowley’s (1996) conceptual model, however, the subdivision of scale is different, with the scale component being broken down into the building, block, street and district levels, since the towns and cities are already mixed at that scale—although insufficiently perhaps. This chapter divides the mixed-use development into four scales: buildings, streets, communities and cities, and mainly focuses on the community scale.

Our study area is Southern Changping, which is a typical urban fringe area in

Beijing (Figure 15). It is planned as the new development zone of Beijing since 2004

(http://zhengwu.beijing.gov.cn/ghxx/ztgh/t833101.htm), with the mission to undertake the residential and industrial functions used to be performed in the central city.

Meanwhile, the accelerated development in this area has caused severe urban problems.

Under this background, mixed-use development is set as one of the development strategies to tackle these problems of rapid development in Southern Changping.

Figure 15. Study area and case study spots.

Three types of communities, which are developed respectively through centrally- controlled development model, individual-dominant development model and collective- dominant development model, are studied in this chapter (Table 5). All three communities are mixed-use developed communities.

Type Development Sampling Area Characteristics Model Community in “top-down” Yunquyuan and Large-scale residential Large-scale centrally- Longyueyuan zones when first built; Residential controlled residential other functions are Clusters (LRC) development communities in gradually introduced in model Huilongguan and recent years Tiantongyuan county

UV near the “Bottom-up” Four-north Villages Near the planned TOD individual- (including functional-mixed TOD dominant Shigezhuang, East center; developed by development Banbidian, West individual villagers model Banbidian and Dingfuhuangzhuang)

Village- “Bottom-up” Zhenggezhuang Independent collective collective- Village urbanization; led by the Agent dominant village collective Community development committee (VAC) model Table 5. Three types of communities and development models

(1) Communities in Large-scale Residential Clusters (LRC): This type of community is centrally planned through the “top-down” process under the control of

Beijing and Changping government, represented by residential communities in

Huilongguan (Figure 16) and Tiantongyuan located at the Southern Changping. These communities were built in the early 21st century by the local government as a bedroom community to provide economically affordable housings for the immigrants and the low-

70 or middle-income families. With the population explosion, however, problems appeared in these single-function residential communities, so the mixed-use development is included in recent planning and constructions.

Figure 16. Communities in Large-scale Residential Clusters (LRC).

(2) Urban Villages near the TOD (UV): Urban villages represent the “bottom-up” individual-dominant development model. The case studied in this chapter is the Four- north Villages located in the Huilongguan sub-district in Southern Changping. In 2008, a

TOD center was established by Changping government as a mixed-functional shopping mall that combines commercial, recreational, official and residential functions. However, the development of TOD triggered the construction booms in surrounding villages. As a result, individual villagers in the nearby Four-north Villages (Figure 17) built their own

71 houses spontaneously without legal or regulatory approval, and then rent the spare rooms in their houses to the low-income class and immigrants who could not afford the houses in the government-built residential areas. The build of residential houses gave rise to the construction of infrastructures, retails, education centers and entertainment facilities, and the community soon became the urban villages embedded in the city. Since the urban villages are developed neither led by the government nor by a community committee, the development process is quite dispersive, resulting in the irregular landscape, disgusting environment and chaotic social management.

Figure 17. Urban Villages.

(3) Village-collective Agent Community (VAC): As a representative of “bottom- up” collective-dominant development approach, this community is planned and developed under the lead of the collective committee. Zhenggezhuang Village (Figure

18) is chosen as a model case in this chapter. The village has made use of the rural collective-owned construction land, to establish a modern community that includes residential areas, village-owned industries, entertainment facilities, public spaces and universities. In the development process, all detailed policies, planning and development decisions are decided by the villagers’ committee on the premise of not violating the city master planning. In just 20 years, this community has successfully achieved the “bottom- up” independent urbanization.

Figure 18. Village-collective Agent Community (VAC).

A land-use mix index is adopted here to depict the urban form characteristics of the three communities. The measure of land-use mix follows (Larry Frank et al, 2004;

2006):

LUM = - Σpi ln pi / ln n,

where pi is the percentage of land use area devoted to land use i, and n is the number of land uses. The seven land uses used to calculate this measure were residence, industry, commerce & finance, education & research, green field, water and other. Table

6 shows the percentage of different land uses to total community area and the land mixed index calculated, which reveals that all three communities are mixed-use communities, while the degree of land-use mix of LRC is lower than the other two, with a large proportion of land being built as residential areas.

LRC UV VAC Percentage of Different Land Uses (%) Residence 66.44 6.9 19.36 Industry 12.5 2.69 0 Commerce & Finance 7.63 3.14 3.43 Education & Research 3.82 23.01 16.59 Greenland 5.92 10.5 7.19 Water 0 50.66 43.51 Other 3.69 3.1 9.93 Land-use Index LUM 0.637 0.728 0.844 Table 6. Build environment of three communities

4.3.2 Data

Data are collected by on-the-spot questionnaire survey with proportional stratified sampling. Respondents are randomly picked at several spots of the community (spots are chosen based on pre-interviews and discussion with community leaders to guarantee the coverage and representativeness of respondents). Due to the special characteristics of urban villages (residents refrain from going outside due to the higher crime rate), we also 75 conduct the household surveys on urban villages. Despite the inevitable bias of sampling

(eg. self-selection, healthy user bias, etc.), our sampling is still representative of our research goal. The survey was conducted during the period of April, 2012 to July, 2012.

279 effective questionnaires have been obtained in total (effective rate is 94.9%) with a share of 199, 35 and 45 for LRC, UV and VAC, which accounts for about 5‰ of the total population in each community.

4.3.3 Methods

Based on interviewees’ subjective feelings, we conducted a systematic evaluation to compare the urban performance of the three approaches, from the perspective of job- housing pattern, residents’ career development, sense of community and community vitality.

(1) Job-housing Pattern: We calculated the distance-based and time-based workplace-residence separation rate. The distance-based separation rate is the percentage of residents whose one-way commuting distance exceeds 15 kilometers, while the time- based separation rate is the percentage of residents whose one-way commuting time exceeds 70 minutes1.

(2) Residents’ Career Development: In our research, we investigated residents’ annual income in 2011 as well as their income change during the past three years, to measure residents’ career development in three communities.

1 The 15 kilometers and 70 minutes are the calculated average commuting distance and time in our study area. 76

(3) Sense of community: Sense of community is examined by eight questions concerning class identification, social attachment and the relationship between work and living space. Residents’ reflection is measured using Likert Scales, with numbers from 1 to 5 represents “Completely disagree”, “A little disagree”, “Neutral”, “A little agree” and

“Completely agree”.

(4) Community Vitality: Community vitality is studied through the frequency of five interactive activities in the community, including three professional-oriented activities, one social-oriented activity and one relative-oriented activity. The frequency is measured by Likert Scales, too, with number from 1 to 5 represents “Never”, “Unusual

(once or twice a year)”, “Sometimes (once or twice a month)”, “Often (once or twice a weed)” and “Always (more than twice a week)”.

One-way analysis of variance (ANOVA) is applied to analyze the difference of sense of community and vitality in the three communities. It is a tool to compare the means of different sample of groups. Duncan Method is used for items with the homogeneity of variance, and Games-Howell Method for those without homogeneity of variance. The level of significance test is 5%.

4.4 Results

4.4.1 Positive dimension: urban performance

(1) Job-housing Pattern

Workplace-residence separation is a key indicator in modern urban studies.

Mixed-use development can reduce workplace-residence separation rate by including necessary functions in a community, thus to ease the traffic pressure in large cities. Table

7 shows the comparison of workplace-residence separation rate among the three communities. The distance-based separation rate is the percentage of residents whose commuting distance exceeds 15 kilometers; the time-based separation rate is the percentage of residents whose commuting time exceeds 70 minutes. We can conclude that the separation rate in LRC is the highest, which reveals that in the centrally- controlled developed communities, the compensatory endeavor to infuse mixed functions can hardly address the workplace-residence separation problem; the individual development tends to build residential communities close to the TOD center for living convenience, thus the separation rate in UV is lower than the LRC; VAC has the lowest distance-based and time-based separation rate, mainly benefits from the implementation of mixed-use development from the very beginning of community development.

LRC UV VAC Distance-based Workplace-residence Separation Rate 67.18% 58.62% 52.27% Time-based Workplace-residence Separation Rate 52.31% 41.38% 40.91% Table 7. Comparison of workplace-residence separation rate

(2) Residents’ Career Development

A truly functional mixed-use community is not only supposed to provide a decent housing for its residents, but more importantly, serve as an effective career incubator that helps promote career development of residents living in the community. We focused on the annual income and income growth to test the career development of residents in the three communities. As is shown in Figure 19, residents in UV have the lowest annual income, and the residents with the middle-level income in the community (25th to 50th quantile) suffer from income decline between 2008 and 2011 (Figure 19b). This reveals the poor career development of residents in urban villages, and is caused by several reasons, such as income inequality in large cities, relocation of residents after income increase due to the poor living conditions in urban villages, and the low-end industries and services in this community.

Residents in LRC have the highest income, especially the top 5% residents, whose annual income is significantly higher than those in the other two communities. The income of residents in VAC enjoys the greatest increase in three years, which implies a powerful potential for career development for residents in this community.

Figure 19. The comparison of residents’ annual income.

[(a) Quantile chart of personal annual income of three communities in 2008 and 2011; (b)

Quantile chart of log personal annual income of three communities in 2008 and 2011; (c)

Income change between 2008 and 2011 in three communities.]

(3) Sense of Community

One-way ANOVA analysis is used to compare sense of community. Before the

ANOVA analysis, homogeneity test of variances is conducted in order to choose 80 appropriate ANOVA method. Results (Table 8) show that the Significance (Sig.) of most items is greater than 0.05, indicating the homogeneity of variances on 95% significance level, thus Duncan method is used for these items. Games-Howell is chosen for the item

“Many residents in my community have business relation with me”, as its Sig Value is less than 0.05. The ANOVA results are shown in Table 9. As we can see, sense of community is low in all three communities, yet some differences do exist.

Levene Sig. Sense of community Items Statistic “Residents in my community are of the same class as me” 1.850 0.159 “There are many friends in my community” 0.523 0.593 “There are many peers of the same occupation in my community” 0.169 0.844 “Many residents in my community have business relation with me” 6.249 0.002 “My conversations in community involve topics about my work” 1.707 0.183 “My friends in community has helped me with my difficulties in work” 0.643 0.526 “My community has positive influence on my career development” 0.384 0.681 “My professional circle has nothing to do with my community” 1.947 0.145 Table 8. Results of homogeneity test of variances in the sense of community analysis

Residents in UV and LRC have higher class identification, indicating the homogenization of population composition in these two communities: most residents in

LRC are middle-income class, while immigrant workers constitute the majority in UV.

By contrast, the population in VAC is more diverse, with different classes such as villagers, immigrants, workers, and even some experts working for the university and industrial park in the community.

With regard to the social attachment, UV is significantly higher than the other two communities. Community forms and residents’ characteristics can explain this difference.

On the one hand, the residents in UV have to share the kitchen and restroom, which largely increase the chance to make new friends in the community; on the other hand, most residents in UV are new arrivals in Beijing, who will be more likely to choose the same community as their acquaintances.

In our study, particular attention was paid to the relationship between working and living space. Recent urban development in Southern Changping has moved in the opposite direction to the planned goal as “improving the technology, education, R&D and industrial functions”, as the residential functions far exceed the employment function under the influence of the booming housing market in China. Therefore, we try to evaluate whether the residential communities could serve as an incubator for industrial development and employment. For the five questions that measure the relationship between working and living space, only the answers to “There are many peers of the same occupation in my community” have significant difference among three communities, that the UV plays a more positive role in the occupational promotion of residents. As a result, we conclude that the working and living space are isolated in modern communities, while UV is a relatively effective incubator for industrial development.

To conclude, residents in UV have higher class identification and better interpersonal relationships. However, the main reason for this high sense of community is that most residents belong to the low-income class, so the high attachment is achieved at 82 the cost of development. On the other hand, the analysis of the sense of community reveals the unsustainability of centrally-controlled development process represented by

LRC, since the social stratification and lack of public space in this type of community are barriers for community integration and inclusion.

Table 9. One-way ANOVA analyzing results of the sense of community

Communities Sense of community F P Sig. LRC UV VAC Value Value A. Class identification “Residents in my community are 1:0.076 3.191,2 3.292 2.871 2.579 0.078 of the same class as me” 2:0.601 B. Social Attachment “There are many friends in my 1:0.671 2.721 3.372 2.821 4.696 0.010 community” 2:1.000 C. Relationship between Working and Living space “There are many peers of the 1:0.789 same occupation in my 2.471 2.892 2.521,2 2.523 0.082 2:0.064 community” 0.381 “Many residents in my 0.587 community have business 2.051 2.201 2.311 1.639 0.196 0.088 relation with me” “My conversations in 1:0.060 community involve topics about 2.641 3.061 2.801 2.326 0.100 my work” “My friends in community has 1:0.194 helped me with my difficulties in 2.391 2.691 2.561 1.273 0.282 work” “My community has positive 1:0.166 influence on my career 2.431 2.401 2.711 1.326 0.267 development” “My professional circle has 1:0.068 nothing to do with my 3.411 2.941 3.381 2.051 0.131 community” 83

[Mean value in each community is provided following each item, and the superscripted numbers show the result of difference: same superscripted numbers mean that residents in these communities gave a similar evaluation to this question, while different numbers mean that their answers are significantly different. F Value, P-Value and Sig. show the result of the significance test. When P<0.05, the three patterns are significantly different, and vice versa. Games-Howell Method gets three Sig for each question, showing separately the pairwise difference between every two communities from the three. And the Sig of Duncan Method shows the similarity of patterns in the particular group. When

Sig<0.05, the two approaches are significantly different, and vice versa.]

(4) Community Vitality

The loss of vitality is considered one of the most crucial problems in functional zoning. Therefore, vitality is an important indicator for the effectiveness of mixed-use development. We use the frequency of five types of interaction in communities to depict the community vitality.

According to the results of homogeneity test of variances shown in Table 10, Duncan

Method is used for items “Meet with peers of same occupation” and “Meet with old schoolmates”, while Games-Howell Method is applied to the other four items.

Levene Statistic Sig. Community Interactions “Meeting with colleagues” 4.244 0.015 “Meeting with peers of same occupation” 0.164 0.848 “Meeting with service subscribers” 5.832 0.003 “Getting to know new friends” 7.013 0.001 “Meeting with relatives” 6.100 0.003 “Meeting with old schoolmates” 1.620 0.200 Table 10. Results of homogeneity test of variances in community vitality analysis

We can see from the results of ANOVA analysis shown in Table 11, that residents in the three communities have a similar frequency of the socially-oriented interaction

(“getting to know new friends”) and the friendship-oriented interaction (“meeting with old schoolmates”). However, residents in VAC have a higher frequency of the relative- oriented interaction (“meeting with relatives”), since most villagers are still living in the same community.

The three communities show a significant difference in professional-oriented interactions, surveyed by the frequency of residents to meet with their colleagues, peers of same occupations and service subscribers. As we can see, the frequency of interactions in LRC is the lowest, revealing the lack of community vitality in this type of community.

Community Interaction Communities F P Sig. LRC UV VAC Value Value A. Professional-oriented Interactions 0.074 “Meeting with colleagues” 2.641 3.231,2 3.332 5.407 0.005 0.952 0.039 “Meeting with peers of same 1:0.349 2.541 3.202 2.801,2 3.524 0.031 occupation” 2:0.136 0.031 “Meeting with service 1 2 1,2 0.547 subscribers” 1.71 2.37 1.95 4.884 0.008 0.405

B. Social-oriented Interactions 0.243 “Getting to know new 1 1 1 0.140 friends” 3.06 3.51 2.91 2.640 0.073 0.708

C. Friend & Relative-oriented Interactions 0.782 “Meeting with relatives” 1.881 1.741 2.522 4.950 0.008 0.027 0.037 “Meeting with old 1:0.235 2.061 2.261 2.381 1.268 0.283 schoolmates” Table 11. One-way ANOVA analyzing results of the community vitality in community

[Mean value in each community is provided following each item,, and the superscripted numbers show the result of difference: same superscripted numbers mean that residents in these communities gave a similar evaluation to this question, while different numbers mean that their answers are significantly different. F Value, P-Value and Sig. show the result of the significance test. When P<0.05, the three patterns are significantly different, and vice versa. Games-Howell Method gets three Sig for each question, showing separately the pairwise difference between every two communities from the three. And the Sig of Duncan Method shows the similarity of patterns in the particular group. When

Sig<0.05, the two approaches are significantly different, and vice versa.]

4.4.2 Normative dimension: development process

The comparison of the communities under different development models is summarized in Table 12 and Figure 20.

Job-housing Residents’ Career Sense of Community Pattern Development community Vitality LRC * **** ** * UV **** * **** *** VAC ***** *** ** **** Table 12. Comparing results of three communities

[The more “*” represents the higher effectiveness.]

Figure 20. The radar chart of the three types of communities

From the results, LRC, the “top-down” centrally-controlled developed community, suffers from a low sense of community and vitality, as well as a very high workplace-residence separation rate, which will lead to urban problems such as traffic jam, landscape devastation and the loss of urban vitality. Therefore, the dominant “top- down” centrally-controlled development may cause some problems in mixed-use development. Firstly, central planning tends to give preference to the uniformity of landscape, which will lead to functional-zoning. Secondly, the “top-down” development only relies on the decision-making of a small group of elites and planners. As cities are the most complex adaptive systems (CAS), there are explosively growing factors influencing urban development, and it is especially difficult for such a tiny group of elites to identify all these factors and to keep a balance between different stakeholders. What is more, the process of central planning is rather lengthy, making it practically impossible to quickly adjust the development to the fast-changing world in the era of diversity, variability and uncertainty. Admittedly, the “top-down” development can be well implemented in mixed-use development in some cases when appropriate approaches are conducted. However, the problems of this development model merit great attention when being applied to urban development.

Urban villages that come into being through “bottom-up” individual-dominant development are quite common in large cities. Beneath the surface of arbitrary constructions and poor living conditions, these communities are in some ways reasonable from the perspective of mixed-use development, as individual residents tend to establish various functions in their communities to improve their own living convenience.

Consequently, urban villages have advantages in the job-housing pattern, sense of community and vitality. However, while realizing mixed-use development, the communities under individual-dominant development will confront some other urban problems, such as the stagnant career development of residents and the low-quality living conditions in the community. Besides, the individual-dominant development always fails to comply with the mandates of central planning, so the communities built under this model are considered the “grey zone” in cities, and are facing the fate of demolition. As a result, the opportunity of urban villages to achieve mixed-use development is often overlooked in the process of demolition and relocation. For these reasons, in a practical application, the “bottom-up” individual-dominant development often results in arbitrary constructions and the inconformity with central planning, thus does not have the opportunity to achieve effective mixed-use development at most of the time.

Compared to the other two development models, the “bottom-up” collective- dominant development is the most effective one to achieve mixed-use development.

Through collective-dominant development, Zhenggezhuang village not only completed the independent urbanization, but successfully built a modern community with the mixed functions of residence, office, commerce and entertainment on the premise of conforming to the central planning, and achieves the benefits of mixed-use development in job- housing pattern, career incubator, sense of community and vitality. The success of VAC reveals the effectiveness of collective-dominant development.

Firstly, the collective-dominant development encourages all stakeholders to play a positive role in the community development, and keeps a balance between the 89 requirements of different social classes and groups. For example, Zhenggezhuang Village was developed through the Village-Enterprise Jointly Development, that the villagers’ committee, villagers and community industries all participate in the development process either directly or indirectly: villagers entrust the villagers’ committee to help them achieve their aspirations; villagers’ committee acts a leading role, and is in charge of the planning, decision-making and management of the whole community; companies and industries in the community is managed by Hongfu Group, a village-owned corporation, and a certain part of their profit will be used to develop the community; Hongfu Group helps with the development decisions of the community, and provides the financial support for the village (Figure 21). The benefits of the community development are shared by the stakeholders by means of joint-stock, stockholding system and welfare insurance. Secondly, the development dominated by the small villagers’ committee could adjust immediately according to the change of market and residents’ requirement. Take

Zhenggezhuang Village as an example, when the residents were in urgent need of a modern community, the villagers’ committee immediately gathered the village-owned land and established a new community; to guarantee the source of income for the community development, a batch of effective industries and companies were quickly introduced into the village; with the trend of community upgrading in the 21st century, the community further developed commercial, tourism, entertainment and educational industries. As a result, the Zhenggezhuang Village completed the development of a mixed-use modern community in 20 years by itself. Last but not least, the existence of

90 collective agents can avoid the chaos in construction, and ensure the consistency of planning concerning central planning and the urban development trend.

Figure 21. The village-enterprise joint development of Zhenggezhuang village

Apparently, the collective-dominant development model is not a panacea. In addition to the doubts about the legality of this “bottom-up” approach itself, some scholars and planners have also questioned about the sustainability of the collective- dominant development. Unlike the “top-down” planning, the main resource of the development dominated by the collective entity comes from inside the community, and once the resources run out, say, the breakdown of collective-own industries, the whole community will get into trouble. Also, the collective-dominant development depends on a powerful collective entity to gather resources and organize all stakeholders to work on their own planning and construction. If the collective entity such as the villagers’ committee is not strong enough, the development will fall into disorder as individual-

91 dominant development. Due to these reasons, the development approach of

Zhenggezhuang Village is hard to be replicated by other communities. However, the

“bottom-up” collective-dominant development carried out by Zhenggezhuang Village is instructive in mixed-use development, that future urban planning should encourage the

“bottom-up” development and the “trial and error” process of collective entities.

4.5 Discussions and Conclusions

Different urban forms generally lead to different urban performance. In cities across the globe, the compact urban form of mixed-use development has been proved to be an effective strategy to address urban problems caused by functional division and to achieve the goals of urban development as envisioned in “new urbanism” and “smart growth” strategies, and is considered as a significant step towards creating and maintaining the good urban performances such as vital, beautiful, just, environmentally benign human settlements (Talen, 2005). However, previous experiences in many parts of the world have shown that cities have different performance even with the similar urban form characterized by compact and mixed-use. In many Chinese cities, mixed-use development has been practiced in recent decades, but most of these practices just simply fuse various functions into an area without adequate consideration of the mechanism and development approaches, resulting in quite different urban performances.

In this chapter, we try to demonstrate that different development process will also lead to different urban performances in mixed-use development. This chapter assessed three models commonly used in modern urban development in China: the “top-down” 92 centrally-controlled development, the “bottom-up” individual-dominant development and the “bottom-up” collective-dominant development. We found that the mixed-use communities developed through these three processes have quite distinct performance.

Therefore, the effective urban performance requires not only urban forms, but more importantly, the proper development process to realize the urban forms, functions and performance. The key to achieving a successful mixed-use community should be the development process, other than simply plan and build the compact city forms.

Using three communities in Southern Changping of Beijing as a case study, this chapter evaluated the effectiveness of the three development models. Given the complication of Chinese cities and communities due to its great disparities in geographic setting, historical trajectories, cultural background and socio-economic development, we should be cautious to make generalizations based on the case study of only three communities. However, the disparate urban performance of the three communities resulting from different development models reported in this chapter does shed new light on the practice of mixed-use development. Our results show that both UV and VAC that developed bottom-up have advantages in job-housing balance, sense of community and vitality over the top-down developed LRC, which implies that the dominant centrally- controlled development is in some ways less effective in achieving the goals of mixed- use. In the two communities developed by “bottom-up” approach, the UV of individual- dominant development can achieve mixed-use community, but the chaos and inconformity with central planning offset the benefits of mixed-use development. The

“bottom-up” collective-dominant development represented by VAC can effectively

93 achieve the mixed-use development in urban areas, but the lack of strong central administration might threaten the sustainability of the future development. Therefore, the

“bottom-up” development is necessary when carrying out mixed-use development, but this approach can be effective only when combining with the “top-down” development.

Mixed-use development is not simply the mechanical mix of different functions, but a negotiated process of different interest groups (Saich & Hu, 2012). No development model itself is perfect to achieve the effective mixed-use development. Only by combining the “top-down” and “bottom-up” development models, and encouraging the broadly social inclusion and individual participation in the development process, can the desired benefits mixed-use development really be realized.

Chapter 5: Network Perspective - Integrating SCD Analysis with Transportation Design

Despite the rapid increase of private car ownership in China during past two decades, public transportation remains still the most popular means for commuting among urban residents in most Chinese cities. The provision of adequate urban transportation is a challenge for transportation planners. This paper focuses on the design of bus stop distribution based on bus smart card data (SCD) analysis in Shenzhen, China.

Design of bus stop distribution must trade off two components of accessibility in the provision of services: the physical access to transit services (requires more stops) and the geographic space reachable within a given travel time budget (requires fewer stops).

Considering this tradeoff, I propose a Multi-Criteria Evaluation (MCE) approach to quantitatively assess the service supply of the bus system, and derive the travel demand of citizens from the bus SCD and taxi trip data. From the supply and demand, the gap and redundancy of bus supply could be identified, and based on which, a geodesign-based heuristic optimization is conducted to design the bus stop location in order to minimize the service gap and redundancy.

5.1 Research Context

Considering the problems associated with the highly competitive but strongly automobile-dominated urban transportation, such as congestion and pollution, promoting 95 alternative modes of travel, especially public transit, has become an important component of transportation planning (Taaffe et al., 1996). Public transit is considered to have advantages regarding transportation efficiency, social equity, and environmental sustainability. With the ability to meet the requirements of high-volume travel demand, public transit is an irreplaceable mode in urban areas, especially in high-density areas, to meet travel demand and release congestion issues (Pucher, 2004). Socially, public transit is supposed to serve the population (Bullard, 2003), especially the disadvantaged groups like low-income population and elder people (Dodson et al., 2004). Another distinguishing positive characteristic of public transit is its relatively lower energy consumption and fewer pollutant emissions per passenger mile than private automobiles

(Garrison & Levinson, 2006).

The attraction of public transit implies the importance of the design of public transportation system to provide potential users with an alternative to travel which can be competitive with automobiles. In China, specifically, public transit provision is of extreme importance, as a large number of citizens rely on public transit for their daily commute. For instance, in 2015, more than half of commuters in large Chinese cities, such as Beijing (Beijing Transport Institute, 2016), Shanghai (Li, 2015), and Shenzhen

(Hu, 2015), use public transportation. As an important mode of public transportation, the bus system plays a key role in people’s daily commute in Chinese cities. Therefore, the design of a bus system is especially critical in China to attract passengers and satisfy daily commuting.

Many attempts have been applied to bus system design, including the design of bus stop locations (Chien & Qin, 2004; Ibeas et al, 2010; Delmelle et al., 2012), timetable

(Ceder & Wilson, 1986), service frequencies (Jansson, 1980; Ceder, 1984), fleet size

(Jansson, 1980; Glaister, 1986), and bus routes (Bielli et al., 2000). This chapter focuses on the design of bus stop distribution, as it is especially important in optimize the existing bus system. The key to bus stop location design is to address the spatial mismatch between transportation supply and travel demand. One way to address the spatial mismatch of bus stop locations is to expand the stop coverage by adding more stops

(Ibeas et al., 2010). Adding bus stops will improve the accessibility to the bus system, but slow down travel speeds and reduce the reachable destinations within a travel time budget (Murray & Wu, 2003). Another way to improve bus stop design is to achieve faster travel speeds by reducing redundant stops (Murray & Wu, 2003). This strategy will improve the efficiency of the bus system, but reduce the coverage of the system of potential riders. This trade-off between access and efficiency is essential in transportation planning, and will be further illustrated in Section 5.2.1.

The assessment of services provision by current bus system is the first step of bus stop distribution design. Accessibility of transportation services is a popular and well- established indicator in bus system evaluation and planning. There is extensive research on the accessibility of transportation in general, and bus system in particular, from both temporal and spatial perspective. The majority of research on temporal accessibility focuses on the simulation of estimation of travel or waiting time in bus travel (Miller,

1999; Hawas, 2013; Farber et al., 2014). Spatial accessibility indicators can be

97 categorized into physical access (Hsiao et al., 1997; Zhao et al., 2003; Biba et al., 2010) or geographic space ones (Dong et al., 2006; Gutierrez & Garcia-Palomares, 2008;

Delmelle & Casas, 2012), corresponding to the trade-off between access and efficiency in transportation planning. In this chapter, a multi-criteria evaluation (MCE) approach is adopted to assess the service supply of current bus system, to incorporate both the access and efficiency factors in accessibility measures.

After evaluating the current bus supply, optimization is a widely-used approach to design the locations of bus stops. Existing research on optimizing bus stops locations has been focused on determining the optimal spacing of bus stops along bus routes (Saka,

2001; dell’Olio et al., 2006), reducing bus stop redundancy (Delmelle et al., 2012), and minimizing the costs of the transportation systems (Chien & Qin., 2004; Ibeas et al.,

2009). In this chapter, we design the locations of bus stops with the goal to balance bus supply and travel demand, by minimizing the gap and redundancy in the service supply.

Specifically, the goal of this chapter is to design bus stop locations that can better balance the bus service supply and citizen’s travel demand. To achieve this, we firstly conduct an assessment of the current bus system, by measuring the service supply using

MCE approach and determining the travel demand based on the transportation Origin-

Destination (OD) data. Then, the gap of service provisions is identified and redundant stops are recognized. Based on the gap and redundancy of bus supply, scenarios of bus stop locations are generated using a semi-empirical approach based on GIS, and a heuristic optimization of bus stop locations is conducted to minimize both the gap and redundancy. The case study is conducted in Shenzhen, China. This chapter contributes to 98 the current research in the following aspects: (1) it proposes a MCE approach to evaluate the bus service supply regarding the trade-off between access and efficiency in transportation planning; (2) it provides a framework to identify the gap and redundancy in the current transportation systems; (3) it proposes a geodesign-based heuristic optimization approach for bus stop locations design, by integrating the geospatial analysis with design process and applying a heuristic algorithm to generate design scenarios and optimize the stop locations.

The rest of the chapter is organized as follows: section 5.2 discusses the trade-off in public transit planning, followed by the discussions of previous literature on bus system assessment and optimization and the research framework of this chapter that corresponds to this trade-off; section 5.3 introduces the data and methodology used in this study; section 5.4 reports the results of bus system assessment and bus stops locations design; and finally, section 5.5 contains the conclusions and the discussions of future research.

5.2 Transportation Design: Gap and Redundancy

5.2.1 Accessibility trade-off in public transit planning

For bus system planning, there are two aspects of accessibility that are of particular importance: physical access and system efficiency (Murray & Wu, 2003). The first aspect, the physical access to transit services, can be measured by bus stop coverage.

A 5 min or 400m walking distance is considered a reasonable access standard for bus 99 stop transit in urban areas (Demetsky & Lin, 1982; Levinson, 1983; Fitzpatrick et al.,

1996; Schöbel, 2005). The more bus stops that exist, the higher the coverage, and the greater the access. However, more stops mean slower travel speeds of the buses, resulting in less system efficiency, which is the second aspect of the system. The system efficiency refers to the geographic space that is reachable from a particular location within a given travel time budget (Murray & Wu, 2003). Regarding system efficiency, fewer stops along a route will increase travel speeds and broaden the geographic space that can be reached over a fixed period of time (Fitzpatrick et al., 1996; Furth & Rahbee, 2000; Delmelle et al., 2012). However, decreasing the number of stops will decrease the access to transit services by passengers (Murray, 2003). This trade-off exists between the number of transit facilities, and striking a balance between these two aspects is a critical challenge in transit planning (Murray & Wu, 2003; Delmelle et al., 2012).

Considering the trade-off between ‘adding more stops for higher access’ and

‘reducing stops to improve system efficiency’, a reasonable bus stop spacing is a critical issue in bus system planning. Previous research has put efforts on establishing standards for the spacing of transit stops. Ammons (2014) states that stop spacing standards of 200–

600m for bus systems are common, while Demetsky and Lin (1982) argues that spacing in some areas can reach 800m. However, considering the heterogeneous demographic characteristics in urban areas, the travel demand always varies across space, stop spacing should also be adjusted across space to meet the diverse travel demands.

Therefore, to address the trade-off in bus stops locations design, the following issues should be fully considered: (1) the measure of bus service supplies should include 100 both the physical access and system efficiency components; (2) the heterogeneous travel demands across space should be measured to determine bus stop spatial distribution. In this chapter, a MCE approach is adopted to measure bus supply considering both access and efficiency factors, which will be illustrated in section 5.2.2. The travel demand will be derived from the individual-level bus smart card data (SCD) and taxi trip data, to capture the spatial heterogeneity of travel demands.

5.2.2 Bus service assessment: a MCE approach

This chapter adopts a multi-criteria evaluation (MCE) approach to measure bus service supply that considers both access and efficiency. MCE is a very common research method for transit performance assessment (Hassan et al., 2013; Hawas et al., 2016), aiming at handling the trade-offs among multiple objectives, and at quantifying and ranking performances based on relative scores. MCE transforms the numerical values of indicators into a 0-1 scale with 0 representing the worst choice and 1 the best.

There are several well established MCE methods, among which the AHP

(Analytical Hierarchy Process) proposed by Saaty (1980) and TOPSIS (Technique of

Order Preference by Similarity to Ideal Solution) put forward by Hwang and Yoon (1981) are most commonly used. The AHP (Saaty, 1980) method establishes the weights of criteria based upon the pairwise comparisons. On the other hand, TOPSIS method

(Hwang & Yoon, 1981) is used to compare scores of options (eg. different spatial units of the study area) according to an individual criterion or aggregate using all criteria.

TOPSIS estimates the so-called ideal and ideal-negative alternatives based on the values 101 and weights of the indicators. More specifically, this method firstly identifies the ‘ideal’ value among all the spatial units in the study area (the unit with the highest value among all studies units), and then the score of each unit is then calculated by normalizing the value of itself against the ‘ideal’ value. After this normalization, the score of the criterion is a real number between 0 and 1, where 1 indicates an ideal performance and 0 means the poorest performance. This method is very efficient in identifying the relatively deficient areas, and keeps the internal consistency among different criterion (as the values of all criteria are ranging from 0 to 1).

TOPSIS method is used in this study to measure the values of several criteria related to transit supply, and to calculate the score of bus service supply by aggregating all criteria. The most commonly-used indicators of transit supply in previous research include service coverage (Polzin et al., 2002; Currie, 2004; Hawas et al., 2016), service frequency (Ryus et al., 2000; Polzin et al., 2002; Currie, 2004; Fu & Xin, 2007; Curtis &

Scheurer, 2010; Hawas et al., 2016), travel/waiting time (Polzin et al., 2002; Currie,

2004; Fu & Xin, 2007), and route coverage (Fu & Xin, 2007; Hawas et al., 2016).

Referring to the criteria used by Hawas et el. (2016), this chapter adopts three indicators

(stop coverage, service frequency, and route diversity) to measure access, and use number of opportunities reachable over a certain time budget to measure efficiency

(Table 13). Literature has indicated the use of different approaches for the weighting process to reflect the importance of different evaluation criteria (Yeh et al., 2000; Wang

& Lee, 2009; Hassan et al., 2013; Hawas et al, 2016). The weighting is usually determined based upon the opinions from a group of unbiased transportation experts

102

(Hassan et al., 2013; Hawas et al., 2016). In this study, however, due to the logistical and time constraints and the limited scope of this study, we do not undertake a survey of experts to determine the weight of these factors considered through either the AHP or

TOPSIS method. Therefore, we choose to use equal weight (0.5) to the access and efficiency criterion, to balance these two factors. Urban planners could flexibly change the weights of criteria based on their specific design target or their investigation when applying our method to their own cases. Among the indicators measuring access, stop coverage is a measure of the spread of the transit service, and is determined by whether the area lies within a buffer zone of a bus stop; service frequency is a measure of the

‘magnitude’ of transit service in a district, and is represented by the number of scheduled bus pick-ups within a time period at each bus stop. Two indicators are used to measure service frequency: trips per person measures the scheduled bus pickups per person, and trips per area measures the scheduled bus pickups per area unit; route diversity represents the access to and from different regions via bus services, and is measured by the number of bus stops than can be reached directly (without any transfer) or with one transfer. To balance the different indicator, we apply equal weight to the indicators of access (0.1).

For opportunity indicator, point of interest (POI) is used to represent urban opportunities, and the number of POIs reachable within a time budget is calculated to measure the efficiency of the bus system. Greater weights are assigned to the number of POI reachable within a shorter time budget, which is more in accordance with the reality.

103

Table 13. Supply criteria and indicators

Criterion Indicator Definition of indicator (of district i, Indicator Criterion weight i=1,2,…,n) weight Whether the area lies within a 400m Stop buffer zone of bus stops coverage 퐶푖 0.1 (Ci) 1, i is in the buffer zone = { 0, i is not in the buffer zone Trips in the time period per person: total scheduled bus pickups in district 0.1 Service i/total population in district i frequency Trips in the time period per square Access 0.5 (F ) i kilometer area: total scheduled bus 0.1 pickups in district i/area of district i Direct destination: number of bus stops that can be reached from district i 0.1 Route without any transfer diversity One-transfer destination: number of (R ) i bus stops that can be reached from 0.1 district i with only one transfer Number of POI (points of interest) that can be reached from district i in 15 0.2 minutes Number of POI (points of interest) that Opportunity Efficiency 0.5 can be reached from district i in 30 0.2 (O ) i minutes Number of POI (points of interest) that can be reached from district i in 45 0.1 minutes

The selected criteria and indicators reach an excellent balance between the access and efficiency factors in public transit planning. To increase stop coverage, service frequency, and route diversity, more bus stops are required, while to increase opportunity, fewer bus stops are necessary to reduce travel times.

104

5.2.3 Bus stop design: a geodesign-based heuristic optimization approach

In the design of transportation systems, heuristics and metaheuristics methods, such as Generic Algorithm (GA), Ant Colony (AC), and Simulated Annealing (SA), are always used to get nearly optimal solutions efficiently for practical sized problems

(Farahani et al., 2013). These methods are proposed based on analogies to physical, chemical, or biological processes. However, one deficiency of these methods is that they are relatively efficient in computational speed, and they usually need to be run on a supercomputer with very high computing capability. This makes it hard to apply these methods to urban design and the high-cost in repeating the optimization process..

This chapter adopts a simpler geodesign-based heuristic optimization approach.

Geodesign is the thought process comprising the creation of entities in our geo-scape, or more simply, the design in geographic space (Miller, 2012). The key characteristic of geodesign is that it brings geographic analysis into the design process, provides a design framework and supporting technology to combine geography with design (Dangermond,

2009). Reflecting on the ideas of geodesign, in this chapter, instead of generating scenarios and optimize the bus stop locations from pure computational and mathematical process, we design the bus stops according to the spatial distribution of bus supply gap and redundancy. The empirical results from MCE analysis are used to generate scenarios about the new bus stops that should be built to fill in the supply gap and the existing stops that could be removed to reduce supply redundancy. Based on the design goal to address the spatial mismatch between bus service supply and travel demand, our optimization goal is to minimize the gap and redundancy of bus supply. To obtain the optimal solution, 105 a heuristic algorithm is processed to iterate through all design scenarios to generate the optimal bus stop distribution.

5.3 Study Area, Data, and Methodology

5.3.1 Study area

The study area of this chapter is Shenzhen, China. The municipality covers an area of 2,050 square kilometers including urban and rural areas, with a total population of

14.5 million in 2011. About 55% of daily trips in this city are dependent on the public transit system (9.9 million trips currently). There are over 830 bus lines in Shenzhen and bus trips account for 67.5% of daily public transit trips (http://www.sztb.gov.cn/). The bus system serves nearly 7 million passengers per day.

In 2004, Shenzhen introduced a smart card program and installed relevant equipment on buses (Shi & Lin, 2014). Smart cards can be used on all the bus lines and metro in Shenzhen. When boarding onto and alighting from the bus, passengers holding such cards can tape cards to the card readers. However, passengers only need to tape cards once when boarding if bus lines have the same price (not distance-based). When a smart card is placed within the range of a card reader, smart card transaction details (the card number, boarding time, bus machine ID as well as other operational data) are created and stored in the machine. The development of smart card technologies provides an opportunity for researchers to analyze the data for transport planning.

106

Considering the scope of this chapter, we only evaluate and design the bus stops locations to meeting the demand of commuting during morning peak hours (7-9am on working days). There are around 300,000 trips during this time period. Therefore, the design of bus stops locations is very important to meet the demand of commuting.

5.3.2 Data

The following data are used in this chapter:

(1) Bus smart card data (SCD): the SCD during 7-9am on January 12-16, 2015 are used to derive the service frequency, direct and one-transfer destinations, and destinations reachable within a certain time budget. The data is also used to measure travel demand.

The original data includes the boarding time and coordinates information for each bus ride. The alighting time and coordinates, boarding/alighting stops, and bus routes of each ride are derived.

(2) Taxi trip data: the taxi trip data during 7-9am on January 12-16, 2015 are used to measure travel demand. The data includes information of every take-on and drop-off times and locations.

(3) Bus stops and routes: the data were crawled from amap (http://m.amap.com/), which is a popular online map in China (similar to Google Map).

(4) POI data from Weibo: Sina Weibo is a popular social media used by Chinese citizens (similar to Twitter). In this chapter, we used the POI (point of interest) crawled from the Weibo to determine the number of urban opportunities reachable in a certain

107 time budget. In total, 34,898 POIs are identified using the Weibo data crawled during

January 2014 to February 2015.

5.3.3 Methods

(1) Supply measures: A MCE approach

As described in section 5.2.2, the bus supply in the study area is measured through a MCE approach, using indicators stop coverage, service frequency, route diversity, and opportunity. The supply measure is conducted over a 30*30m raster layer, and the supply score Si of each cell i (i = 1, 2, ..., n) is calculated.

Stop coverage (Ci) is measured by whether the cell i falls in the 400m buffer zone of the bus stops. Ci is 1 when cell i is in the buffer zone, and is 0 when it is not in the buffer zone.

Service frequency is measured by the number of total trips per person or per square kilometers. The number of trips of each bus stop is calculated by Algorithm 1. In this algorithm, L denotes the set of all bus routes; l is a certain bus route and l ∈ L ; S denotes the set of all bus stops; s is a certain bus stop s ∈ S; R denotes the set of all bus boarding records; r is a certain boarding record r ∈ R; Fi denotes the total trips of bus stop si.

108

Algorithm 1 Service Frequency based on Bus Smart Card Data 1: for each bus route l ∈ L do 2: for each bus stop s of route l, s ∈ S and s ∈ L do 3: sort all boarding records according to boarding time 4: record the first boarding time t0 5: for each boarding record of the stop r ∈ R do 6: if the next boarding time is less than 5 minutes later of the record t0 then 7: delete the boarding record 8: else 9: keep the boarding record and change t0 to this boarding time the number of boarding records is the service frequency of stop s 10: find the stop smax with the greatest service frequency 11: for stops before smax [s0, s1, …, smax] do 12: if service frequency Fi-1 of stop si-1 is smaller than service frequency Fi of stop si then 13: if boarding time of stop si is later than 7:10am then 14: Fi-1 – Fi 15: else 16: Fi-1 = Fi - 1 17: for stops after smax [smax, …, sn] do 18: if service frequency Fi+1 of stop si+1 is smaller than service frequency Fi of stop si then 19: if boarding time of stop si is earlier than 8:50am then 20: Fi+1 – Fi 21: else 22: Fi+1 = Fi - 1

Route diversity is measured by the total number of destinations that can be reached from cell i either directly or with only one transfer. Algorithm 2 is used to derive the set of all stops that can be reached directly (D0) and the set of all stops that can be reached in one transfer (D1).

109

Algorithm 2 Route Diversity (calculate the number of direct destinations and destinations with one transfer) 1: for each bus stop s ∈ S do 2: D0 = stops on the same bus line as s 3: for each bus stop s ∈ D0 do 4: L1 = bus lines that s ∈ L1 5: D1 = bus stops on all bus lines in L1

The efficiency of the bus system is measured by the ‘opportunity’ indicator, which counts the total number of POIs that can be reached by bus during a travel time budget (15/30/40 minutes). The destinations and POIs reachable are derived using

Algorithm 3.

Algorithm 3 Opportunity (calculate the number of POIs reachable in time budget T) 1: for each bus stop s ∈ S do 2: Rs = alighting records with time interval t ≤ T 3: A = all unique alighting stops in Rs A ⊆ Rs 4: for each alighting stops a ∈ A do 5: Oa = all POIs in the 400m buffer zone of stop a

The supply scores of each indicator are then standardized to 0-1 scale using the minimum-maximum normalization method. The standardization makes the scores of different criteria comparable and additive to calculate the overall supply measure.

Finally, let i denotes the cell index of the n cells in the study area, then i = 1, 2,

…, n; let j denotes indicator j of J total indicators, then j = 1, 2, …, J; Ij is the relative score of indicator j, and Wj is the weight of indicator j, then the overall supply measure of cell i is calculated by:

110

퐽 푆푖 = ∑푗=1 푊푗퐼푖푗, (1)

And the total supply of the study areas is calculated by:

푛 퐽 푆 = ∑푖=1 ∑푗=1 푊푗퐼푖푗, (2)

(2) Demand measures

The travel demand is derived from the bus SCD and taxi trip data. The total number of boarding of each bus stop is derived from bus SCD, and is assigned to the buffer of the bus stop. For taxi trip data, two cases are considered: (1) if the taxi take-on takes place within the buffer zones of bus stops, then the travel demand is assigned to the buffer; (2) if the taxi take-on occurs outside the buffer zones of bus stops, then a 400m buffer zone is created for this taxi take-on location and the travel demand is assigned to this buffer, as a new bus stop located in either location of this 400m buffer zone could be accessible to this potential rider. After obtaining the travel demand of the whole study area, the demand in each cell is then standardized to 0-1 scale using the same minimum- maximum normalization method.

(3) Gap and redundancy

After obtaining the supply scores and demand scores of each cell, a supply- demand indicator of each cell is calculated by:

푆퐷푖 = 퐷푖 − 푆푖 , (3)

SDi is an indicator of bus supply gap and redundancy. For cells with SDi > 0, which indicates that the supply is smaller than the demand, there is a gap in this area and more bus stops might be needed. On the contrary, if SDi < 0, which indicates that the 111 supply surpasses the demand, then the area has redundant bus service supply, and urban planners could consider removing the bus stops in these areas to improve the efficiency of the bus system.

(4) Bus stop locations: heuristic optimization

Based on the empirical measures of bus supply gap and redundancy, a heuristic algorithm is adopted to find the optimal distribution of bus stops. Firstly, based on the spatial distribution of supply gap and redundancy, scenarios are generated regarding which existing stops are redundant and could be removed and where the new stops should be built to fill in the gap. Then, the scenarios are iteratively tested through a heuristic algorithm (Figure 22) The design goal of locating new stops is to minimize the supply gap, while the goal of removing redundant stops is to minimize the supply redundancy.

The score of the overall gap G is calculated as:

푛 G = ∑푖=1 푆퐷푖, 푆퐷푖 > 0, (4)

and the score of the overall redundancy R is calculated as:

푛 R = ∑푖=1|푆퐷푖| , 푆퐷푖 < 0, (5)

112

Figure 22. Heuristic algorithm of bus stop location optimization

In the optimization, the order of scenarios being optimized (the order of stops being added/removed) matters to the final result. So the optimization program is run repeatedly to figure out the correct order of scenarios, so that the stops that will reduce

113 the gap/redundancy to the greatest extent will be tested first. This makes the design more efficient, as the change that will make greater different should be considered first.

5.4 Results

5.4.1 Data processing and OD matrix estimation

Bus origin-destination (OD) data are crucial for urban mobility study and bus system planning. Collecting such data, however, is extremely difficult. The traditional way to obtain OD data is by conducting travel survey on passengers, which is time- consuming and labor-intensive, and can only obtain a small pool of sample compared to the large commuting population. A more efficient way to gather bus data is to utilize smart card data (SCD) generated by the automatic fare collection (AFC) system. The

SCD capture rich spatial and temporal information of each passenger through the smart cards held by each passenger, thus can greatly save the time and labor in the data collection process.

Thanks to its large volume and high spatial-temporal resolution, bus SCD is playing an increasingly essential role in urban studies. For example, SCD analysis can help obtain a better knowledge of user commuting behavior, as each user’s spatial and temporal pattern during his/her trip can be generalized (Agard et al., 2006; Bagchi &

White, 2005). Previous work related to SCD analysis mainly focuses on four aspects: smart card data processing and Origin-Destination (OD) matrix estimation (Agard et al.,

2006; Morency et al., 2007; Trépanier et al., 2007; Munizaga & Palma, 2012), transit operation and management (Trépanier & Morency, 2010), commuting behavior and 114 travel patterns (Ma et al., 2013; Park et al., 2008; Zhong et al. 2015; Zhou & Long,

2013), and spatial structure of cities (Liu et al., 2012; Zhong et al., 2014).

Recent work has also raised the importance and attractiveness of using SCD for transportation planning. Bagchi and White (2005) advocated the complementation of

SCD in relation to other data collection methods in transportation studies and planning, and conducted three case studies in British transportation networks to illustrate the application of SCD. Utsunomiya et al. (2006) studied transit ridership and advocated different bus schedules for different days based on the variability in rider demand. Deakin and Kim (2001) used SCD to compare actual and planned network to provide users with an alternative itinerary before a journey. Among all research, network optimization has been widely applied as a method to conduct transportation planning with a specific target.

Researchers have developed a variety of models or methods to optimize bus stop spacing

(Furth & Rahbee, 2000; Alterkawi, 2006; Chien et al., 2004; Delmelle et al., 2012), route design (Mandl, 1980; Pattnaik et al., 1998) and the whole network layout (Yang et al.,

2007) to improve transportation accessibility.

Bus SCD are originally generated to facilitate digital payment, public transport ticketing and bus system management (Pelletler et al., 2011; Kurauchi & Schmöcker,

2017). Unfortunately, such data are often lack of information such as the boarding stop, alighting time and alighting stop, which makes the application of the data to academic research and urban planning difficult, because knowing both the origin and destination of each individual trip is critical. Therefore, data processing is crucial to make the smart card data ‘smarter’ to be used in urban planning.

115

To process the bus SCD and enable its application in urban transportation studies, the following workflow is adopted to estimate the OD matrix (Figure 23). There are two phrases in the data process. The first phrase is boarding information extraction, which includes: (1) space match between bus AFC data and bus GPS data to derive the boarding locations; (2) bus line match to obtain the bus line information of each SCD record;

(3)travel direction extraction that determines the direction of each record by comparing

GPS locations and bus stop locations. After completing all the above steps, we have all the ‘O’ (origins) information required by the OD matrix. The goal of the second phrase is to derive the ‘D’ (destinations) information. An approach based on Transition-Attraction

Rules (TAC) is proposed to estimate the alighting stops and times for each boarding record.

116

Figure 23. Data processing and OD matrix estimation of bus SCD

(1) Space match To understand the mobility trajectory of a passenger, it is useful to obtain the boarding time and location of each SCD record. As the bus AFC data do not provide boarding location information, we need to conduct space match to obtain the coordinates of each boarding record. This matching could be achieved with the help of the bus GPS data, which record the location of each bus every several seconds. The method is to 117 compare boarding time with the GPS time, and if the time in two datasets is close enough to each other, the GPS coordinates are assigned as the boarding location (Figure 24a). As the example shows in Figure 24b, the boarding time (T*) is very close to the GPS time t5, then the GPS coordinates at time t5 will be the location of the boarding record. The matching rate of this process is about 90%, while the 10% failure of matching is mainly due to the missing of some GPS data.

Figure 24. Space match: (a) space match by comparing boarding time with GPS time (b) an example of space match

(2) Bus line match In the bus line match process, we need to match the route id in the bus SCD (a set of unique ids) with the bus lines in the real-world bus system (Figure 25a). Instead of selecting bus lines that match with each bus SCD record, an ‘inverse selection by probability’ method based on geographic analysis is proposed to compare and match the routes recorded to the real bus line system (Figure 25b). For each bus line, a buffer is generated, and the probability of each route (a set of recorded point locations) that falls into the buffer is compared and the route with the highest probability will be selected as

118 the route that matches with the bus line. This method will significantly increase the match speed while at the same time maintain a high successful matching rate (about 95%).

Figure 25. Matching bus routes and bus lines through inverse selection by probability (a) bus line match between two datasets (b) an example of inverse selection by probability

(3) Direction extraction Many buses in Shenzhen is traveling in two directions, so it is necessary to determine the direction of each bus. Previous methods are trip-based, that the direction is determined by analyzing the set of GPS coordinates of every bus trip. However, this method requires a clear separation of every bus trip, which is not a simple task as some buses will not stop after arriving at the destination of the previous trip and will start a new trip right away. Considering this issue, the method we developed is geography- based. Instead of analyzing all the GPS records of a complete bus trip, our method focuses on each boarding record, and extract the direction by comparing the GPS

119 coordinates before and after the boarding with the locations of previous and next bus stops. As is illustrated in Figure 26, the bus could travel in two directions, then we have: direction =

1, d(퐺푃푆 , 푆푡표푝 ) + d(퐺푃푆 , 푆푡표푝 ) < d(퐺푃푆 , 푆푡표푝 ) + d(퐺푃푆 , 푆푡표푝 ) { 푝푟푒 푝푟푒 푛푒푥푡 푛푒푥푡 푝푟푒 푛푒푥푡 푛푒푥푡 푝푟푒 (1) −1, d(퐺푃푆푝푟푒, 푆푡표푝푝푟푒) + d(퐺푃푆푛푒푥푡, 푆푡표푝푛푒푥푡) > d(퐺푃푆푝푟푒, 푆푡표푝푛푒푥푡) + d(퐺푃푆푛푒푥푡, 푆푡표푝푝푟푒)

Figure 26. Orientation extraction methods

(4) Alighting estimation Bus AFC system only records boarding information, so we need to estimate the alighting time and location. To obtain such information, I propose a Transition-Attraction

Rule (TAR) approach for alighting inference.

The TAR approach consists of two parts. The first part is the alighting inference based on transition rules. For each passenger with multiple trips per day, the trip chain is generated and the alighting stops will be inferred based on his/her transition between different buses. Estimation based on transition rules will return relatively accurate results, but can only be applied to passengers with full trip chain (whose travel track is a full

120 circle) and passengers who transfer between different buses. For the rest of records, attraction rules will be used. Gravity model will be applied to measure the attraction between bus stops and the alighting stops will be estimated based on the attraction. This estimation can obtain results of nearly all SCD records, but the estimated alighting information is not as accurate as the one estimated by transition rules.

Transition rules apply to passengers with more than one boarding records in one day. It is assumed that for every single trip, the passenger gets off the bus at the nearest stop of his/her next boarding; and for the last trip of the day, the passenger returns to his/her origin of the day (Figure 27).

Figure 27. Alighting inference based on transition rules

In this alighting inference model, two constraints are applied to improve the accuracy of the estimation. The first constraint is the distance constraint that requires the alighting location of the previous trip to be close enough to the boarding stop of the next

121 trip, so that the passenger is able to walk to the next boarding stop for bus transition.

Then we have:

′ 2 ′ 2 √(푥2 − 푥1) + (푦2 − 푦1) ≤ 푑푡ℎ푟푒푠ℎ표푙푑 , (6)

′ ′ where we assume the alighting location of the first trip is (푥1, 푦1), and the boarding location of the next trip is (푥2, 푦2). The distance threshold (푑푡ℎ푟푒푠ℎ표푙푑) represents the distance constraint between previous alighting stop and next boarding stop.

The distance threshold is set to be less than 1km in most previous studies, but as we have another constraint to exclude some unreasonable alighting inference, the threshold in this study is set as 3km. A larger threshold value can avoid the exclusion of some extreme situations (e.g. long-walking trip).

Another constraint is the time constraint, which guarantees that the passenger has enough time to travel from the first boarding stop to the next boarding stop before the bus arrives. The constraint is represented as:

푡푏푢푠 + 푡푤푎푙푘 ≤ 푡2 − 푡1， (7)

where 푡2 − 푡1 is the time interval of the two consecutive boarding records of the passenger, 푡푏푢푠 is the time for the passenger to travel from the first boarding stop to the first alighting stop by bus, and 푡푤푎푙푘 is the time for the passenger to walk from the first alighting stop to the next boarding stop, estimated by the distance between two stops and the average walking speed (1.25 m/s). Instead of using the threshold as time constraint, the time interval between two consecutive boarding (푡2 − 푡1) is adopted in our approach, to include both short-time bus transfers and long-time activities between two bus trips.

For example, a passenger might go to the supermarket on his/her way home, so the 122 transfer time interval will be relatively long. If we use time threshold value other than the time interval as constraint, this transition will be considered unreasonable and the trip will be neglected.

Therefore, for each passenger that has several bus trips recorded in one day, we first determine the nearest bus stop on the bus line of his previous trip to his next boarding, and if the bus stop meets both the distance and time constraints, it will be assigned as the alighting stop of the previous trip. For the last trip of each passenger in the day, only distance constraint is considered, as it is the final destination of the passenger.

For the rest of the records, attraction rules are applied. The logic of this rule is to calculate the probability of each passenger to alight at a bus stop based on the attraction between that stop and the boarding stop. Gravity model is adopted to measure the attraction between stops. The model could be described as follows:

For every boarding stop 푠푏, there is a set of possible alighting stops, including all the stops that this bus will pass through:

S={푠1, 푠2, … 푠푛} (8)

For every possible alighting stop 푠푖, we can calculate the attraction force between it and the boarding stop 푠푏 using the gravity model:

−훽 퐴푖 = 푘푃푏푃푖⁄(푑푏푖 ) , (9)

where 푃푏 is the population of the boarding bus stop, 푃푖 is the population of the alighting bus stop 푠푖, 푑푏푖 is the distance between stop 푠푏 and 푠푖, β is the distance decay factor, and k is a constant. 푃푏 and 푃푖 could be estimated using Weibo check-in data. A set 123 of attraction forces between boarding stop and all possible alighting stops can then be obtained:

A={퐴1, 퐴2, … 퐴푛} (10)

The probability of alighting at stop 푠푖 could be calculated based on the attraction forces:

푛 푝푟표푏푖 = 퐴푖⁄∑1 퐴푖 (11)

And the set of probability of all possible alighting stops could be obtained:

prob={푝푟표푏1, 푝푟표푏2, … 푝푟표푏푛} (12)

With all the probabilities, a random number r (between 0 and 1) could be generated and the alighting stop could be assigned based on the value of r:

푠1, 0 < 푟 ≤ 푝푟표푏1 2 푠2, 푝푟표푏1 < 푟 ≤ ∑1 푝푟표푏푖 alighting stop = 푠 , ∑2 푝푟표푏 < 푟 ≤ ∑3 푝푟표푏 (13) 3 1 푖 1 푖 … 푛−1 푛 {푠푛, ∑1 푝푟표푏푖 < 푟 ≤ ∑1 푝푟표푏푖

After conducting above data processing, the OD matrix of the study area is obtained and ready to be used for further analysis.

5.4.2 Positive dimension: gap and redundancy of bus supply

(1) Supply measures

Using the MCE approach, we measure the bus service supply of the study area from both the access and efficiency perspective (Figure 28). The measure of access to the bus system takes into consideration the bus stops coverage, the frequency of scheduled bus pick-ups, and the destinations that can be reached directly or within only one transfer.

124

The standardized access scores of some major roads (e.g. G107, G205, Shennan Street,

Songbai Road, Jihua Road, etc) are relatively higher, indicating the citizens have higher accessibility to the bus system around these major roads. To our surprise, the access of the bus system in the southern Shenzhen, where is the CBD of Shenzhen and have high- density population, does not appear to be higher than the other areas. The efficiency score, on the other hand, distributes very uneven across space. The efficiency of the bus system in the southern and southwestern Shenzhen is much higher than in the other areas.

The western, eastern, and northern Shenzhen have very low efficiency score, meaning that the bus systems there are operating in lower efficiency, with lower travel speed and fewer POIs that can be reached. However, besides the relatively faster travel speed of buses, the high efficiency in the southern Shenzhen can also be attributed to the relatively more POIs in this area, as this is the CBD of Shenzhen and have more urban activities and facilities. This fact also reveals the potential deficiency of the methods that only considers efficiency factors in accessibility or supply measures.

The measure of bus service supply using MCE approach balances the access and efficiency factors (Figure 28 bottom). The results indicate the higher supply not only in the southern and southwestern areas, but also around the major roads in other areas, which is more in accordance with the reality.

125

Figure 28. Supply of the bus services in Shenzhen from the access (top) and efficiency

(middle) perspective, and the overall supply measure (bottom). 126

(2) Demand measures

The travel demand of citizens is derived from bus SCD and taxi trip data (Figure

29). The travel demand based on bus SCD is dispersedly distributed across space. An interesting finding of this demand measure is that most of the demands locate around the activity centers, rather than inside the center areas. For example, in the southern

Shenzhen where there is a dense street network, most travel demand distributes around periphery areas of the CBD. The reason is that during the morning peak hours, most travel demand is from the periphery areas to the activity centers. This phenomenon implies the potential gap of bus supply in the periphery areas, as these areas tend to have relatively lower supplies. The travel demand derived from the taxi trip data is more aggregated. The demand is most densely distributed in the southern Shenzhen, and is also aggregated in the western, northern, and northeastern areas. Combining the travel demand derived from the two data source, we obtain the overall travel demand of the citizens in

Shenzhen (Figure 29 bottom).

127

Figure 29. Travel demand of the citizens in Shenzhen from the bus SCD (top) and taxi trip data (middle), and the overall measure of demand (bottom).

128

(3) Gap and redundancy

The distribution of gap and redundancy in the study area is then assessed by comparing the supply and demand. The supply gap distributes dispersedly in the city

(Figure 30), and there are relatively more severe gaps in the southern part of the city near the CBD areas. More closely investigation shows that the gaps have a banding distribution in the northern and middle of the city along the western-eastern direction, and a northern-southern banding distribution along the Fulong Street.

Figure 30. Gaps of bus supply in Shenzhen (top) and the stripes with concentrated gaps

(bottom) 129

The distribution of supply redundancy, on the other hand, is highly concentrated

(Figure 31). Most redundancy gathers in the southern Shenzhen, and also has the banding distribution. This result is reasonable, as in the city center where there is dense population and urban activities, urban planners tend to distribute more facilities, which might result in supply redundancy.

Figure 31. Bus supply redundancy in Shenzhen

5.4.3 Normative dimension: design of bus stop locations

Based on the supply gap and redundancy, we generate the scenarios of the bus stop locations (Figure 32). In total, 328 design scenarios are generated, with 151 stops to be established and 177 stops to be removed.

130

The scenarios of the new stops that could be built are generated based on the supply gap, with the goal to reduce the gaps in bus service supplies. The scenarios are generated in the center of the gap clusters, but is located as close as possible to the streets. The locations of existing stops are also considered when generating new design scenarios, to avoid locating new stops too close to the existing stops, so that the new stops will not increase supply redundancy.

Considering the areas with supply redundancy, we propose that urban planners could consider removing some stops, so that the resource could be more efficiently used.

If an existing bus stop locates at areas with high redundancy (SDi ≤ -0.4), then the stop is considered to be removed. The stops to be removed are selected carefully by considering the distribution of bus stops: only the stops that locate very close to the other stops will be put into the scenario. This control makes sure that the remove of stops will not cause too much trouble to citizens’ commute, as they can choose a nearby stop without too much additional walking. Also, the ‘remove’ of a bus stop does not necessarily mean that this stop will be deleted. In the real world planning, urban planners could choose to reduce the scheduled bus pickups at these stops or to replace the lines with several stops to be removed by express lines that stop less.

131

Figure 32. Scenarios of the bus stops being built and removed

The new design scenario of bus stop locations is then generated from optimization

(Figure 33). In total, 89 stops are designed to be added and 47 stops are to be removed.

The new stops are distributed dispersedly across the city, and are located along the major roads that have higher gaps. The redundant stops are concentrated in the southern area of the city where there are more redundant services. After the optimization, the new bus stop distribution achieves less gap and redundancy (Figure 34).

132

Figure 33. Design of bus stop locations

Figure 34. The gap and redundancy of the new design scenario

To evaluate the bus supply of the new design scenario, the overall gap (G) and redundancy (R) of the entire system is calculated using formula (4) and (5). It is obvious that the overall gap and redundancy has decreased after the optimization (Table 14). We also calculated the percentage of cells that have gaps and redundancy in service supply, 133 and results show that the cells with gap have reduced by about 50%, and the cells with redundancy have reduced by about 20%.

Gap Redundancy Gap-score (G) % of gaps Redundancy-score (R) % of redundancy Current system 2335.59 5.07% 4730.42 5.54% New system 2257.13 1.12% 2033.68 4.49% Table 14. Improvement of the bus service supply after optimization

We also plot the change of gap (Figure 35) and redundancy (Figure 36) with the number of stops being added and removed. The gap/redundancy reduce sharply with the

first several stops being added/removed, and then the change becomes less significant.

Urban planners can use these plots to determine the number of stops to be added/removed

according to their policy goals.

Change of Gap 2340 2330 2320 2310

2300 score

- 2290

Gap 2280 2270 2260 2250 2240 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 Number of stops being added

Figure 35. Change of gap as the number of stops being added

134

Change of Redundancy 5000

4500

4000 score - 3500

3000

2500 Redundancy

2000

1500 1 6 11 16 21 26 31 36 41 46 Number of stops being removed

Figure 36. Change of redundancy as the number of stops being removed

The optimization approach used in this study has the following advantages: firstly, this approach is a semi-empirical optimization, which is based on the empirical analysis of the gap and redundancy of the current system. Compared to the pure numerical optimization algorithm, the empirical analysis makes the optimization results more coincident with the reality and can be applied to transportation planning directly.

Second, the design scenario is generated by adjusting the current system, including adding new stops and remove redundant stops. This process is more in accordance with the planning process in the real-world, which is also based on the adjustment of the current system instead of generating an entirely new layout of the bus network.

135

5.5 Discussions and Conclusions

The goal of this chapter is to design the bus stops locations that can better balance the bus service supply and citizen’s travel demand. The public transit planning has to deal with the trade-off between access and efficiency issues: more bus stops are needed to improve the access to the bus system, while fewer stops should be designed to improve the system efficiency. To balance the access and efficiency factors, we firstly conduct an assessment of the current bus system in Shenzhen, by measuring the service supply using a MCE approach. Results reveal that the major roads have higher access, and the southern portion of the city has higher efficiency. A total supply assessment is conducted considering both the access and efficiency factor, and together with the travel demand derived from the transportation Origin-Destination (OD) data, the gap in service provisions is identified and redundant stops are recognized. The oversupply mainly concentrated in the southern part of Shenzhen, where their population density is high and urban activities are intense. Not surprisingly, planners tend to put more facilities in the southern portion of the city. Regarding the service gaps, the distribution of gaps is more dispersed across the whole city, and most of the gaps have a banding distribution.

Based on the gap and redundancy of bus supply, 328 scenarios of bus stop locations are generated, including 151 stops to be built to minimize the service gap, and

177 stops to be removed to reduce supply redundancy. The final design scenario of bus stop locations is generated using an iterative-optimization process. Compared to the pure numerical optimization process, the method used in this paper is more applicable to the real-world urban planning.

136

This study has some limitations. First, for the MCE approach that assesses the supply of the current systems, equal weights are assigned to the different criteria and indicators. The results may be different if unequal weights could be applied. The weights could be determined by the subjective judgments from experts or by conducting travel surveys. Second, in this chapter, fixed demand is used due to data availability. However, the demand could be dependent on various factors such as bus stop coverage, transportation service frequency, and travel time, and might change as the bus supply varies (Farahani et al., 2013). Therefore, future research could use elastic demand by implementing a model that simulates the change of demand based on some bus system attributes (Chien & Spasovic, 2002). Third, in this case study, we propose that stops could be removed in areas with oversupply. However, this might trigger concerns on social equity issues: can we remove a stop if it serves only one rider? More research could be conducted to deal with the problem of supply redundancy while fully considering social equity.

137

Chapter 6: Summary and Conclusion

6.1 Summary and Conclusions

The primary objective of this dissertation is to build a bridge between data analytics and urban planning for Chinese cities. Inspired by Batty’s (2013) new science of cities, this dissertation aims to integrate the positive and normative dimensions of urban studies. The theoretical framework is developed in Chapter 2/section 2.4 by linking urban analytics with urban design from the space, place, and network perspectives.

In Chapter 3, a geodesign-based CA model is proposed to link the urban simulation with urban design. The model could be used to simulate the patterns of urban growth according to proposed design scenarios, and based on the evaluation of the simulation results, a new design scenario could be generated to better control urban sprawl. The case study conducted in Changping, Beijing demonstrates that the combination of the CA model and geodesign can be beneficial to both the positive and normative dimensions of urban studies: the design serves as the target and guidance of the CA modeling process, and CA models provide the empirical support to a more rigorous design.

In Chapter 4, the traditional field methods are combined with data analytics to study the mechanism and effectiveness of mixed-use development strategy on the place scale. The case study in southern Changping reveals that the bottom-up collective development process is the most effective path to achieve the claimed benefits of the

138 mixed-use development, and the data analytic methods deployed well teased out the mechanism under the implementation of a designed urban development strategy.

The analysis of transportation big data is applied to the design of transportation system in Chapter 5. The transportation big data (e.g. bus smart card data, taxi trip data) records individual travel behavior, and provides a lens to quantify and assess the supply and demand of the current transportation system. By measuring the gap and redundancy of the bus system using the bus smart card data and taxi trip data, we can generate the design scenarios of bus stop locations that are more in accordance with the reality, and derive the nearly optimal design of bus stops distribution in a more straightforward and computational-efficient way.

In addition to the conclusions from the three geographic perspectives, the following conclusions could be drawn from this dissertation:

(1) Theoretically, this study demonstrates that the integration of the positive and normative dimensions of urban studies is feasible and powerful. The normative dimension provides the research goal of the empirical urban studies, and guides the whole empirical process, while the positive dimension provides empirical support for the urban design. The three case studies also have theoretical significances respectively.

Considering the field of urban modeling, the research in Chapter 3 addresses the problem of ‘not application CA models’ by demonstrating an original and powerful application of this type of models. The empirical studies in Chapter 4 contribute to theories concerning urban development, urban form and urban function, and the discussion on ‘top-down’ and

‘bottom-up’ development will be significant to urban planning theories. The research in

139

Chapter 5 deals with the trade-off between the access and efficiency factors in transit planning, and provides the measurement to balance these two factors in the assessment of transportation systems.

(2) Methodologically, the new approach of linking data analytics with urban design is implemented proposed from the three distinct but related perspectives. From the space perspective, an iterative process of conducting urban design through analysis, CA modeling, and scenarios evaluation is put forward; from the place perspective, the development process that can effectively implement mixed-use development strategy is identified through fieldwork and data analytics; from the network perspective, a MCE approach is developed to balance the trade-off between the access and efficiency factors in transit planning, and an iterative-optimization process based on empirical analysis is used for bus stops locations design. These methods are demonstrated to be effective and could be applied to the urban planning in China.

(3) Empirically, this dissertation has local policy implications in the study areas.

The geodesign-based CA model could help policy-makers obtain knowledge concerning urbanization process of Changping, and generate a better land-use design scenario which can effectively control urban sprawl and preserve agricultural land and ecological habitats The comparison of three communities from the place perspective will give a hint to local government about policies concerning urban development and community design, and the methods used to evaluate urban performance provide a paradigm for local government to conduct policy/planning evaluation. The bus stops locations design can be

140 applied to the transportation network planning that will benefit the citizens by providing a more accessible and efficient bus system.

6.2 Future Research

In terms of future studies, I am quite interested in the following topics as immediate follow-up studies:

(1) Uncertainty quantification of the urban models

As stated in Chapter 3, the current CA models are nearly ‘not applicable in the real-world urban design process. To enable broader application of urban models, besides the geodesign framework proposed in this dissertation, the uncertainty of the models should also be quantified and addressed to improve the accuracy of these models in simulating urban growth. The urban big data brings about potentials to reduce data uncertainty of urban models, e.g. we can replace aggregated census data by disaggregated population data derived from cellular signaling data. Besides, the structure uncertainty should also be addressed by refining the model based on urban theories. In this way, we will be able to tackle the growing urban complexity and uncertainty, and develop models to be more suitable and accurate for urban design.

(2) New transportation technologies and future transportation system design

In this dissertation, due to the data constraints, the assessment of the transportation system is based on the traditional transportation modes, such as public transit and taxi. The emerging technologies are generating new transportation modes (eg. ridesourcing, bike sharing), which is bringing about substantial transformations to the

141 mobility patterns of urban inhabitants. Therefore, the design of transportation system should also take into consideration there emerging new transportation modes. The study of supply and demand of these new transportation modes and the change they caused to citizens’ commuting patterns, will significantly influence the design of the public transit systems.

(3) Data analytics and public inclusion

In Chapter 4, we conclude that the bottom-up development process and broader social inclusion is an effective way to achieve the designed benefits of certain urban development strategies. However, how to improve public inclusion in the urban planning process, especially by implementing urban data analytics, remains unanswered. The data analytic methods proposed in this chapter is focused on helping urban planners evaluate the implementation of urban strategies, and the approaches or platforms that can take advantage of urban analytics to encourage/enable social inclusion should be studies to provide technical support to the bottom-up process of urban planning.

(4) Planning support platform development

An impediment to the application of data analytics in urban planning is the high training cost to get urban planners familiar with the analytic tools. Therefore, a user- friendly geospatial platform for applying the urban analytic tools to real-world urban design problems is necessary, so that urban planners will be able to use the platform with simple training. The current planning support systems are seldom been used by urban planners as they seldom integrate the design target with the data analytic process. In this dissertation, we have tested and demonstrated the potential of integrating urban design

142 with empirical data analytic process, and this provides theoretical support to a planning support platform that better combines the design target with the technical tools.

143

References

Agard, B., Morency, C. and Trépanier, M (2006) Mining public transport user behaviour

from smart card data. IFAC Proceedings Volumes, 39(3), 399-404.

Alterkawi, M. M (2006) A computer simulation analysis for optimizing bus stops

spacing: The case of Riyadh, Saudi Arabia. Habitat International,30(3), 500-508.

Ammons, D (2014) Municipal Benchmarks: Assessing Local Perfomance and

Establishing Community Standards. Routledge.

Andy Coupland (Ed.) (1997) Reclaiming the city: mixed use development. Taylor &

Francis.

Bagchi, M. and White, P. R. (2005) The potential of public transport smart card

data. Transport Policy, 12(5), 464-474.

Batty M and Xie Y (1994) From cells to cities. Environment and Planning B 21: s31-s31.

Batty M (1997) Cellular automata and urban form: a primer. Journal of the American

Planning Association 63(2): 266-274.

Batty M, Couclelis H and Eichen M (1997) Urban systems as cellular

automata. Environment and Planning B: Planning and Design 24(2): 159-164.

Batty M, Xie Y and Sun Z (1999) Modeling urban dynamics through GIS-based cellular

Automata. Computers, environment and urban systems 23(3): 205-233.

Batty M (2007) Cities and complexity: understanding cities with cellular automata, 144

agent-based models, and fractal. MIT press.

Batty M (2009) Urban modeling. International Encyclopedia of Human Geography,

Oxford, UK: Elsevier.

Batty M (2013) The new science of cities. MIT Press.

Barnett, J (2007) Smart growth in a changing world. Planning, 73(3).

Beijing-China (2013, Nov. 2) Beijing Master Planning (2004-2020). Retrieved from

http://zhengwu.beijing.gov.cn/ghxx/ztgh/t833100.htm. (in Chinese).

Beiing City Lab (2013) Data 8, Housing price in Beijing. Available at:

http://www.beijingcitylab.com (accessed 10 Jan 2015).

Beijing Municipal Commission of Urban Planning (2005) Beijing Master Planning 2004-

2020, Chapters 4, 5, 6.

Beijing Transport Institute (2016), 2016 Annual report of Beijing Transportation

Development.

Bullard, R. D (2003) Addressing urban transportation equity in the United

States. Fordham Urb. LJ, 31, 1183.

Biba, S., Curtin, K. M. and Manca, G (2010) A new method for determining the

population with walking access to transit. International Journal of Geographical

Information Science, 24(3), 347-364.

Bielli, M., Caramia, M. and Carotenuto, P (2002) Genetic algorithms in bus network

145

optimization. Transportation Research Part C: Emerging Technologies, 10(1),

19-34.

Brail, R. K. and Klosterman, R. E (2001) Planning support systems: integrating

geographic information systems, models, and visualization tools. ESRI, Inc..

Burton, E., Jenks, M. and Williams, K. (Eds.) (1996) The compact city: a sustainable

urban form?. Routledge.

Burton, E (2000) The compact city: just or just compact? A preliminary analysis. Urban

studies, 37(11), 1969-2006.

Calthorpe, P. and Fulton, W (2001) The regional city. Island Press.

Ceder, A (1984) Bus frequency determination using passenger count

data. Transportation Research Part A: General, 18(5-6), 439-453.

Ceder, A. and Wilson, N. H (1986) Bus network design. Transportation Research Part

B: Methodological, 20(4), 331-344.

Chen, H., Jia, B. and Lau, S. S. Y. (2008) Sustainable urban form for Chinese compact

cities: Challenges of a rapid urbanized economy. Habitat International, 32(1), 28-

40.

Chien*, S. I. and Qin, Z (2004) Optimization of bus stop locations for improving transit

accessibility. Transportation planning and Technology, 27(3), 211-227.

Chien, S. I. J. and Spasovic, L. N. (2002) Optimization of grid bus transit systems with

elastic demand. Journal of advanced transportation, 36(1), 63-91.

146

Clarke KC and Gaydos LJ (1998) Loose-coupling a cellular automaton model and GIS:

long-term urban growth prediction for San Francisco and

Washington/Baltimore. International Journal of Geographical Information

Science 12(7): 699-714.

Couclelis H (1985) Cellular worlds: a framework for modeling micro-macro dynamics.

Environment and planning A 17(5): 585-596.

Currie, G. (2004). Gap analysis of public transport needs: measuring spatial distribution

of public transport needs and identifying gaps in the quality of public transport

provision. Transportation Research Record: Journal of the Transportation

Research Board, (1895), 137-146.

Curtis, C., & Scheurer, J. (2010). Planning for sustainable accessibility: Developing tools

to aid discussion and decision-making. Progress in Planning, 74(2), 53-106.

Dangermond J (2009) GIS: Designing our future. ArcNews 31(2): 1.

Deakin, E., & Kim, S. (2001). Transportation technologies: Implications for planning.

Delmelle, E. C., & Casas, I. (2012). Evaluating the spatial equity of bus rapid transit-

based accessibility patterns in a developing country: The case of Cali,

Colombia. Transport Policy, 20, 36-46.

Delmelle, E. M., Li, S., & Murray, A. T. (2012). Identifying bus stop redundancy: A gis-

based spatial optimization approach. Computers, Environment and Urban

Systems, 36(5), 445-455.

147 dell'Olio, L., Moura, J., & Ibeas, A. (2006). Bi-level mathematical programming model

for locating bus stops and optimizing frequencies. Transportation Research

Record: Journal of the Transportation Research Board, (1971), 23-31.

Demetsky, M. J., & Bin-Mau Lin, B. (1982). Bus stop location and

design. Transportation Engineering Journal of ASCE, 108(4), 313-327.

Dodson, J., Gleeson, B., & Sipe, N. G. (2004). Transport disavantage and social status: a

review of literature and methods. Australia: Urban Policy Program, Griffith

University.

Dong, X., Ben-Akiva, M. E., Bowman, J. L., & Walker, J. L. (2006). Moving from trip-

based to activity-based measures of accessibility. Transportation Research Part

A: policy and practice, 40(2), 163-180.

Du, N., &Li, Y. (2005). Planning support system (PSS) and its application to decision-

making for urban spatial development [J]. Journal of Wuhan University of

Hydraulic and Electric Engineering, 1.

Duany, A., Speck, J., & Lydon, M. (2010). The smart growth manual. McGraw-Hill.

Dixon, T. J., & Marston, A. (2003). Mixed Use Urban Regeneration at Brindleyplace,

Birmingham and Gunwharf Quays, Portsmouth: An Assessment of the Impact on

Local and National Economies: Research Report for British Property Federation.

College of Estate Management,| c2003.

Eikelboom T and Janssen R (2015) Collaborative use of geodesign tools to support

148

decision-making on adaptation to climate change. Mitigation and Adaptation

Strategies for Global Change (first published on-line: DOI 10.1007/s11027-015-

9633-4).

Farahani, R. Z., Miandoabchi, E., Szeto, W. Y., & Rashidi, H. (2013). A review of urban

transportation network design problems. European Journal of Operational

Research, 229(2), 281-302.

Farber, S., Morang, M. Z., & Widener, M. J. (2014). Temporal variability in transit-based

accessibility to supermarkets. Applied Geography, 53, 149-159.

Fu, L., & Xin, Y. (2007). A new performance index for evaluating transit quality of

service. Journal of public transportation, 10(3), 4.

Furth, P., & Rahbee, A. (2000). Optimal bus stop spacing through dynamic programming

and geographic modeling. Transportation Research Record: Journal of the

Transportation Research Board, (1731), 15-22.

Fitzpatrick, K., Hall, K., Perkinson, D., Nowlin, L., & Koppa, R. (1996). Guidelines for

the location and design of bus stops. TCRP Report, 19.

Fotheringham, S., & Rogerson, P. (Eds.). (2013). Spatial analysis and GIS. CRC Press.

Frank, L. D., Andresen, M. A., & Schmid, T. L. (2004). Obesity relationships with

community design, physical activity, and time spent in cars. American journal of

preventive medicine, 27(2), 87-96.

Frank, L. D., Saelens, B. E., Powell, K. E., & Chapman, J. E. (2007). Stepping towards

149

causation: do built environments or neighborhood and travel preferences explain

physical activity, driving, and obesity?. Social science & medicine, 65(9), 1898-

1914.

Friedmann, J. (1987). Planning in the public domain: From knowledge to action.

Princeton University Press.

Frey, H. (1999). Designing the city: towards a more sustainable urban form. Taylor &

Francis.

Gardner M (1970) Mathematical games: The fantastic combinations of John Conway’s

new solitaire game ‘life’. Scientific American 223(4): 120-123.

Garreau, J. (2011). Edge city: Life on the new frontier. Random House LLC.

Garrison, W. L., & Levinson, D. M. (2014). The transportation experience: policy,

planning, and deployment. Oxford university press.

Glaister, S. (1986). Bus deregulation, competition and vehicle size. Journal of Transport

Economics and Policy, 217-244.

Grant, J. (2002). Mixed use in theory and practice: Canadian experience with

implementing a planning principle. Journal of the American Planning

Association, 68(1), 71-84.

Grant, J. (2006). Planning the good community: New urbanism in theory and

practice (Vol. 9). Taylor & Francis.

Gutiérrez, J., & García-Palomares, J. C. (2008). Distance-measure impacts on the

150

calculation of transport service areas using GIS. Environment and Planning B:

Planning and Design, 35(3), 480-503.

Handy, S. (2005). Smart growth and the transportation-land use connection: What does

the research tell us?. International Regional Science Review,28(2), 146-167.

Hanzl, M. (2007). Information technology as a tool for public participation in urban

planning: a review of experiments and potentials. Design Studies,28(3), 289-307.

Hassan, M. N., Hawas, Y. E., & Ahmed, K. (2013). A multi-dimensional framework for

evaluating the transit service performance. Transportation Research Part A:

Policy and Practice, 50, 47-61.

Hasse JE and Lathrop RG (2003) Land resource impact indicators of urban

sprawl. Applied geography 23(2): 159-175.

Harris, B. (1960). Plan or projection: An examination of the use of models in

planning. Journal of the American Institute of Planners, 26(4), 265-272.

Hawas, Y. E. (2013). Simulation-based regression models to estimate bus routes and

network travel times. Journal of Public Transportation, 16(4), 6.

Hawas, Y. E., Hassan, M. N., & Abulibdeh, A. (2016). A multi-criteria approach of

assessing public transport accessibility at a strategic level. Journal of Transport

Geography, 57, 19-34.

Healey, P. (1992). Planning through debate: the communicative turn in planning

theory. Town planning review, 63(2), 143.

Herndon, J. D. (2011). Mixed-Use Development in Theory and Practice: Learning from

151

Atlanta's Mixed Experiences.

Hoppenbrouwer, E. and Louw, E. (2005) Mixed-use development: Theory and practice in

Amsterdam's Eastern Docklands. European Planning Studies,13(7), 967-983.

Hsiao, S., Lu, J., Sterling, J., & Weatherford, M. (1997). Use of geographic information

system for analysis of transit pedestrian access. Transportation Research Record:

Journal of the Transportation Research Board, (1604), 50-59.

Hu, G. (2015). Shenzhen public transportation report. Shenzhen Daily Newsletter,

http://www.chinabuses.com/keyun/2015/0924/article_16554.html (accessed on

7/24/2017).

Hwang, C.L, Yoon, K., 1981. Multiple Attributes Decision Making Methods and

Applications. Springer, Berlin Heidelberg.

Ibeas, Á., dell’Olio, L., Alonso, B., & Sainz, O. (2010). Optimizing bus stop spacing in

urban areas. Transportation research part E: logistics and transportation

review, 46(3), 446-458.

Jacobs, J. (1961). The death and life of great American cities. Random House LLC.

Jansson, J. O. (1980). A simple bus line model for optimisation of service frequency and

bus size. Journal of Transport Economics and Policy, 53-80.

Kelliher, D. (1997). The Chinese debate over village self-government. The China

Journal, 63-86.

Klosterman, R. E. (1999). The what if? Collaborative planning support

152

system. Environment and planning B: Planning and design, 26(3), 393-408.

Kurauchi, F., & Schmöcker, J. D. (Eds.). (2017). Public Transport Planning with Smart

Card Data. CRC Press.

Landis J and Zhang M (1998) The second generation of the California urban futures

model. Part 1: Model logic and theory. Environment and Planning B 25: 657-666.

Levinson, H. S. (1983). Analyzing transit travel time performance (No. 915).

Li, J. (2015). Metro passengers will account for around 55% of all commuters in

Shanghai in two years. ThePaper.

http://www.thepaper.cn/newsDetail_forward_1378122 (accessed on 7/24/2017)

Li, Q. (2010). New concepts of community planning: a case study of Playa Vista in USA.

CityPlanning Review, 2010, 34(9): 42-48.

Li X and Yeh AGO (2000) Modelling sustainable urban development by the integration

of constrained cellular automata and GIS. International Journal of Geographical

Information Science 14(2): 131-152.

Li X and Yeh AGO (2002) Neural-network-based cellular automata for simulating

multiple land use changes using GIS. International Journal of Geographical

Information Science 16(4): 323-343.

Linde M, Dias E, and Scholten H J (2015) Geodesign: integrating geographical sciences

and creative design in envisioning a 'New Urban Europe'. International Journal of

Global Environmental Issues 14(1/2): 164-174.

Liu L, Long Y, & Batty M. (2014). A retrospect and prospect of urban models. Beijing

City Lab. Working paper #42 153

Liu, S. (2009). Capitalization of collective land and rural urbanization: A survey of

Zhenggezhuang Village in Beijing. Journal of Peking University, 45(6), 123-132.

(in Chinese)

Liu, Y. (2008). The research on the strategies of urban mixed community. Master

Dissertation of Tongji University. (in Chinese)

Liu, Y., Wang, F., Xiao, Y., & Gao, S. (2012). Urban land uses and traffic ‘source-sink

areas’: Evidence from GPS-enabled taxi data in Shanghai.Landscape and Urban

Planning, 106(1), 73-87.

Long Y, He Y, Liu X and Du L (2006) Planning of the controlled-construction area in

Beijing: Establishing urban expansion boundary. City planning review 12: 004.

Long Y, Mao Q, Shen Z, Du L and Gao Z (2008) Comprehensive Constrained CA

Urban Model: Institutional Constraints and Urban Growth Simulation [J].

In: Urban Planning Forum, Vol. 6, pp. 83-91.

Long Y, Mao Q and Dang A (2009) Beijing urban development model: Urban growth

analysis and simulation. Tsinghua Science & Technology 14(6): 782-794.

Long, Y., Shen, Z. and Mao, Q. (2011). An urban containment planning support system

for Beijing. Computers, Environment and Urban Systems, 35(4), 297-307.

Long Y, Shen Z and Mao Q (2012) Retrieving spatial policy parameters from an

alternative plan using constrained cellular automata and regionalized sensitivity

analysis. Environment and Planning-Part B 39(3): 586.

Long Y and Shen Y (2015a) Data Augmented Design: Urban Planning and Design in

the New Data Environment. Beijing City Lab. Working paper #67

154

Long, Y. and Shen, Z. (2015b). Target or Dream? Examining the Possibility of

Implementing Planned Urban Forms Using a Constrained Cellular Automata

Model. In Geospatial Analysis to Support Urban Planning in Beijing (pp. 19-38).

Springer International Publishing.

Lynch, K. (1984). Good city form. MIT press.

Ma, L. J. (2002). Urban transformation in China, 1949–2000: a review and research

agenda. Environment and planning A, 34(9), 1545-1569.

Ma, X., Wu, Y. J., Wang, Y., Chen, F., & Liu, J. (2013). Mining smart card data for

transit riders’ travel patterns. Transportation Research Part C: Emerging

Technologies, 36, 1-12.

Mandl, C. E. (1980). Evaluation and optimization of urban public transportation

networks. European Journal of Operational Research, 5(6), 396-404.

Miller, H. J. (1999). Measuring space‐time accessibility benefits within transportation

networks: basic theory and computational procedures. Geographical

analysis, 31(1), 1-26.

Miller, W. R. (2012). Introducing Geodesign: the concept.

Morency, C., Trepanier, M., & Agard, B. (2007). Measuring transit use variability with

smart-card data. Transport Policy, 14(3), 193-203.

Municipal Commission of Urban Planning (2007) Changping New Town Planning 2005-

2020.

Munizaga, M. A., & Palma, C. (2012). Estimation of a disaggregate multimodal public

155

transport Origin–Destination matrix from passive smartcard data from Santiago,

Chile. Transportation Research Part C: Emerging Technologies, 24, 9-18.

Murray, A. T., & Wu, X. (2003). Accessibility tradeoffs in public transit

planning. Journal of Geographical Systems, 5(1), 93-107.

Park, J., Kim, D. J., & Lim, Y. (2008). Use of smart card data to define public transit use

in Seoul, South Korea. Transportation Research Record: Journal of the

Transportation Research Board, (2063), 3-9.

Pattnaik, S. B., Mohan, S., & Tom, V. M. (1998). Urban bus transit route network design

using genetic algorithm. Journal of transportation engineering, 124(4), 368-375.

Pelletier, M. P., Trépanier, M., & Morency, C. (2011). Smart card data use in public

transit: A literature review. Transportation Research Part C: Emerging

Technologies, 19(4), 557-568.

Polzin, S., Pendyala, R., & Navari, S. (2002). Development of time-of-day-based transit

accessibility analysis tool. Transportation Research Record: Journal of the

Transportation Research Board, (1799), 35-41.

Pucher, J. (2004). Chapter 8. Public Transportation (pp. 199-236). Guilford.

Qiu B. (2009). Complicated science and urban planning reform [J]. City Planning

Review,2009,04:11-26. (in Chinese)

Rodríguez, D. A., Khattak, A. J., & Evenson, K. R. (2006). Can new urbanism encourage

physical activity?: Comparing a new urbanist neighborhood with conventional

suburbs. Journal of the American Planning Association, 72(1), 43-54.

156

Rowley, A. (1996). Mixed-use development: ambiguous concept, simplistic analysis and

wishful thinking?. Planning Practice and Research, 11(1), 85-98.

Rowley, A. (1998). Planning Mixed Use Development: Issues and Practice.RICS.

Ryus, P., Ausman, J., Teaf, D., Cooper, M., & Knoblauch, M. (2000). Development of

Florida's transit level-of-service indicator. Transportation Research Record:

Journal of the Transportation Research Board, (1731), 123-129.

Saaty, T.L. (1980). The Analytic Hierarchy Process. Mcgraw-Hill, New York.

Saich, T. (2000). Negotiating the state: The development of social organizations in

China. The China Quarterly, 161, 124-141.

Saich, T., & Hu, B. (2012). Chinese Village, Global Market: New Collectives and Rural

Development. Palgrave Macmillan.

Saka, A. A. (2001). Model for Determining Optimum Bus-Stop Spacingin Urban

Areas. Journal of Transportation Engineering, 127(3), 195-199.

Schöbel, A. (2005). Locating stops along bus or railway lines—a bicriteria

problem. Annals of Operations Research, 136(1), 211-227.

Schwanke, D., & Flynn, T. (1987). Mixed-use development handbook. The Institute.

Steinitz C (2012) A framework for geodesign: changing geography by design. Esri.

Straatman B, White R and Engelen G (2004) Towards an automatic calibration procedure

for constrained cellular automata. Computers, Environment and Urban

Systems 28(1): 149-170.

Sui DZ (2011) Rethinking Progress in Urban Analysis and Modeling: Models, 157

Metaphors, and Meaning. In: Yang X (ed) Urban Remote Sensing: Monitoring,

Synthesis and Modeling in the Urban Environment. John Wiley & Sons, pp 371-

382.

Taaffe, E. J., Gauthier, H. L., & O’Kelly, M. E. (1996). Geography of transportation (2nd

ed.). Prentice Hall: New Jersey.

Talen, E. (2005). New urbanism and American planning: The conflict of cultures.

Routledge.

Tobler WR (1970) A computer movie simulating urban growth in the Detroit

region. Economic Geography 234-240.

Tobler WR (1979) Cellular geography. In: Gale S and Olsson G (eds) (2012) Philosophy

in geography (Vol. 20). Springer Science & Business Media, pp 379-386.

Torrens PM (2000) How cellular models of urban systems work (1. Theory). Available

on-line at http://discovery.ucl.ac.uk/1371 (last accessed 02/10/2016)/

Torrens PM and O'Sullivan D (2001) Cellular automata and urban simulation: where do

we go from here?. Environment and Planning B: Planning and Design 28(2):

163-168.

Trépanier, M., & Morency, C. (2010). Assessing transit loyalty with smart card data.

In 12th World Conference on Transport Research, Lisbon, Portugal.

Trépanier, M., Tranchant, N., & Chapleau, R. (2007). Individual trip destination

estimation in a transit smart card automated fare collection system. Journal of

Intelligent Transportation Systems, 11(1), 1-14.

Urban Land Institute. (1987). Mixed-use Development Handbook. Wathington, DC: 158

Urban Land Institute.

Utsunomiya, M., Attanucci, J., & Wilson, N. (2006). Potential uses of transit smart card

registration and transaction data to improve transit planning.Transportation

Research Record: Journal of the Transportation Research Board, (1971), 119-

126.

Wang, T. C., & Lee, H. D. (2009). Developing a fuzzy TOPSIS approach based on

subjective weights and objective weights. Expert systems with applications, 36(5),

8980-8985.

Wang, Y. P., Wang, Y., & Wu, J. (2009). Urbanization and informal development in

China: Urban villages in Shenzhen. International Journal of Urban and Regional

Research, 33(4), 957-973.

Webster, C. J. (1993). GIS and the scientific inputs to urban planning. Part 1:

description. Environment and Planning B: Planning and Design, 20(6), 709-728.

Wei, L., & Yan, X. (2005). Transformation of “urban village” and feasible mode [J]. City

Planning Review, 7, 003.

Weng, L. (1990). Introduction to urban mixed-use development. Master Dissertation of

Tongji University. (in Chinese)

Wheaton, W. C. (2001). Dispersed employment, commuting and mixed land-use in

modern cities. MIT Center for Real Estate.

White R and Engelen G (1997) Cellular automata as the basis of integrated dynamic

regional Modelling. Environment and Planning B 24: 235-246.

Wilson MW ( 2015) On the criticality of mapping practices: geodesign as critical GIS?

159

Landscape and Urban Planning 142(1): 226-234.

Witherspoon, R., Abbett, J. P., & Gladstone, R. M. (1976). Mixed-use developments:

new ways of land use (Vol. 71). Washington, DC: Urban Land Institute.

Wu F and Webster CJ (1998) Simulation of land development through the integration of

cellular automata and multicriteria evaluation. Environment and Planning B 25:

103-126.

Wu F (2002) Calibration of stochastic cellular automata: the application to rural-urban

land Conversions. International Journal of Geographical Information

Science 16(8): 795-818.

Xing, D. (2005). Study on mixed-use in PUD and illuminations to Chinese constructions.

Doctoral Dissertation of Hsinghua University. (in Chinese)

Yang, Z., Yu, B., & Cheng, C. (2007). A parallel ant colony algorithm for bus network

optimization. Computer‐Aided Civil and Infrastructure Engineering,22(1), 44-55.

Yeh, A. G. O. (1999). Urban planning and GIS. Geographical information systems, 2,

877-888.

Yeh AGO and Li X (2001) A constrained CA model for the simulation and planning of

sustainable urban forms by using GIS. Environment and Planning B: Planning

and Design 28(5): 733-753.

Yeh AGO and Li X (2002) A cellular automata model to simulate development density

for urban planning. Environment and Planning B 29(3): 431-450.

Yeh, A. G. O. and Wu, F. (1999). The transformation of the urban planning system in

160

China from a centrally-planned to transitional economy. Progress in

planning, 51(3), 167-252.

Yeh, C.H., Deng, H., & Chang, Y.H.. (2000). Perspectives for practice fuzzy multi-

criteria analysis for performance evaluation of bus companies. Eur. J. Oper. Res.

126, 459-473.

Yin, X. (2007). The research on the design based on mixed-use idea: The analysis of

three examples in the northeast area of Shanghai. Master Dissertation of Tongji

University. (in Chinese)

Ying, S. (2009) Mixed-use experience of land use in North American and England.

Beijing Planning Review. 2009,(2): 110-112. DOI: 10.3969 /j.issn.1003-627X.

2009.02.028. (in Chinese)

Zhao, F., Chow, L. F., Li, M. T., Ubaka, I., & Gan, A. (2003). Forecasting transit walk

accessibility: Regression model alternative to buffer method. Transportation

Research Record: Journal of the Transportation Research Board, (1835), 34-41.

Zheng, S., Long, F., Fan, C. C., & Gu, Y. (2009). Urban villages in China: A 2008 survey

of migrant settlements in Beijing. Eurasian Geography and Economics, 50(4),

425-446. (in Chinese)

Zhong, C., Arisona, S. M., Huang, X., Batty, M., & Schmitt, G. (2014). Detecting the

dynamics of urban structure through spatial network analysis. International

Journal of Geographical Information Science, 28(11), 2178-2199.

Zhong, C., Manley, E., Arisona, S. M., Batty, M., & Schmitt, G. (2015). Measuring

161

variability of mobility patterns from multiday smart-card data. Journal of

Computational Science, 9, 125-130.

Zhou, J., & Long, Y. (2014). Jobs-Housing Balance of Bus Commuters in Beijing:

Exploration with Large-Scale Synthesized Smart Card Data.Transportation

Research Record: Journal of the Transportation Research Board, (2418), 1-10.

Zhu, Y., &Shen, T. (2008). Superplanner: Framework Designing of Planning Support

System Software. Planners, 12, 24-7.

Zhu, X., Wang, Z., & Yin, S. (2010). The evolution and spatial adaptability of mixed-use

settlements: analysis on live-work unit under “Zhejiang Pattern”. Economic

Geography, (6), 933-937. (in Chinese)

Zhuang, S., Ren, L. (2011). Discussions on the mixed-use development strategy of urban

land. Journal of Huazhong University of Science and Technology/Urban

Science, 28(1), 33-37. (in Chinese)

162