URBAN NETWORK ANALYSIS for Rhinoceros 3D
TOOLS FOR MODELING PEDESTRIAN AND BICYCLE TRIPS IN CITIES ANDRES SEVTSUK
City Form Lab HARVARD UNIVERSITY GRADUATE SCHOOL OF DESIGN
URBAN NETWORK ANALYSIS for Rhinoceros 3D
TOOLS FOR MODELING PEDESTRIAN AND BICYCLE TRIPS IN CITIES ANDRES SEVTSUK
City Form Lab
2
CONTENTS � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
4
1 . INTRODUCTION � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
7
FOREWORD �
2.
CASE STUDIES
2.1 .
Predicting walking routes to light rail stops in Surabaya, Indonesia � � � � � � � � � � � � � � � � � � � � � � � 2.1 .1 . The MRT Project � � � � � � � � � � � � � � � � � � � � � � � � � � � � 2.1 . 2. Pedestrian Network Analysis around the Tram � 2.1 . 3 . Feeder streets � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 2.1 .4 . Discussion � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
2. 2.
2. 3 .
3 .
Planning retail centers in Punggol, Singapore � � � 2. 2.1 . Punggol: a transit-oriented residential town 2. 2. 2. The Huff model and modifications � � � � � � � � � 2. 2. 3 . Patronage Analysis � � � � � � � � � � � � � � � � � � � � � � � � 2. 2. 4 . Considering Punggol’s transit-oriented characteristics � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 2. 2. 5 . Simulating new stores � � � � � � � � � � � � � � � � � � � � � 2. 2.6 . Other applications � � � � � � � � � � � � � � � � � � � � � � � �
� � � � �
� � � � � � � � � � � � � � � �
� � � � � � � � � � � �
How Accessibility Explains Retail and Food Establishments' Location Patterns in Cambridge and Somerville, MA � � � � � � � � � � � � � � � � � � � 2. 3 .1 . Study Area � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 2.3.2. Independent Variables: Accessibility Measures 2. 3 . 3 . Results � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 2. 3 . 4 . Applications � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
� � � � � � � � � �
15 15 18 24 25 31 31 32 32 34 37 41
45 45 48 53 55
DOWNLOADING AND INSTALLING
3 .1 . Downloading � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 3 . 2. Installing � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
59 61
3
4 .
GRAPHIC USER INTERFACE
4 .1 .
Manipulating attributes � � � � � � � � � � � � � 4 .1 .1 . Add Text Attribute � � � � � � � � � � � � � 4 .1 . 2. Add Numeric Attribute � � � � � � � � � 4 .1 . 3 . Add Tag Attribute � � � � � � � � � � � � � � 4 .1 . 4 . Remove Attribute from Objects � 4 .1 . 5 . Save Result as Weight � � � � � � � � � � 4 .1 .6 . Show Attribute Tree � � � � � � � � � � � �
4 . 2.
4 . 3 .
4 . 4 .
Import Export functions 4 . 2.1 . Import Points � � � � � � 4 . 2. 2. Import Table � � � � � � � 4 . 2. 3 . Export � � � � � � � � � � � � �
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
Network creation and editing � � � � � � � � � � � � 4 . 3 .1 . Add/Remove Curves from Network � 4 . 3 . 2. Add/Remove Origins � � � � � � � � � � � � � � � 4 . 3 . 3 . Add/Remove Destinations � � � � � � � � � � 4 . 3 . 4 . Add/Remove Observers � � � � � � � � � � � � 4 . 3 . 4 . Delete Network � � � � � � � � � � � � � � � � � � � � Analysis tools � � � � � � � � � � � � 4 . 4 .1 . Accessibility Indices � 4 . 4 . 2. Service Area � � � � � � � � 4 . 4 . 3 . Redundancy Index � � 4 . 4 . 4 . Redundant Paths � � � � 4 . 4 . 5 . Betweenness � � � � � � � � 4 . 4 .6 . Closest Facility � � � � � 4 . 4 .7. Find Patronage � � � � � 4 . 4 . 8 . Distribute Weights � � 4 . 4 .9. Clusters � � � � � � � � � � � �
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
Graphic Options � � � Weight Color �
4 . 5 . 2. Bake
71 71 73 74 77 77 78 79 80 81
83 � 83 � 92 � 94 � 97 � 99 103 106 111 114
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
4 . 5 . Graphics � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � 4 . 5 .1
65 65 66 66 66 67 68
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
5 .
FREQUENTLY ASKED QUESTIONS
Installation � � � � � � � � � � � � Creating networks � � � � � Running UNA analyses
6 .
BIBLIOGRAPHY
119 119 121
� � � � � � � � � � � � � � � � � � � � � � � � � � �
124 125 128
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
133
� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
4
foreword
FOREWORD ANTOINE PICON
G. Ware Travelstead Professor of the History of Architecture and Technology. Harvard University, Graduate School of Design.
Software manuals were at some point considered passÊ because hands-on practice with the software itself was preferred to technical documentation. The Urban Network Analysis user guide represents a shift that has occurred recently in our approach to digital design tools, especially those that deal with the urban scale. It not only offers the reader a technical taxonomy of analytic procedures, but also engages us to reflect on what spatial network analytics actually mean. Such an approach is especially timely for urban planners and designers whose work seeks to promote active mobility in automobile dominated cities. Achieving more walkable, bikeable and sustainable urban mobility requires a better understanding of existing pedestrian and bicycle trip patterns and an ability to predict how these patterns might change when the built environment is changed – via the introduction of new public transit lines, changes in the land use mix or modifications in the spatial fabric of the city. By jointly representing urban form, land uses and mobility flows, the Urban Network Analysis toolbox is part and parcel of a rise in urban modeling tools that reconciliate the study of the physical layout of a city with an understanding of the ways in which it functions. Yet, models are not actual representations of urban reality. They propose ways to interpret a built environments, and above all, frame ways in which we can make decisions and intervene in them. This crucial distinction tends to be often forgotten because of the seductively realistic character of so many digital images. Unlike maps, models also constitute an essential dimension of contemporary urban planning and design decision-making, the power of which should not be underestimated. Through three case studies, this book offers enlightening and nuanced examples of how urban modeling can help shape planning and design decisions in contexts as different as Indonesia, Singapore and the United States.
URBAN NE T WORK ANALYSIS
Implicit throughout this book is the assumption that a city is fundamentally a set of connections between places, in other words a series of networks. Whereas the field of urban planning seems to have endorsed such an approach, urban design still needs practical methods that allow us to relate existing or projected built environments to the ways in which people actually move in them or use them. The Urban Networks Analysis toolbox thus offers a powerful invitation to go beyond traditional urban composition techniques, to distance oneself from the quest for geometric regularity, and to rather develop a deeper understanding of how different elements of the city influence one another, producing patterns of flow and encounter. It makes new urban design solutions possible that go beyond the typically modernist obsession with composition and regularity. Just like the rise of the smart city to which the work contributes, contemporary urban modeling should not be envisaged as a way to restrict design choices – it must be considered as means to inform designs and enhance design flexibility. Such freedom is achieved through what some may perceive as number crunching. Quantitative aspects are indeed fundamental to Urban Network Analysis. But this does not mean that the qualitative dimension of design no longer matters, far from it. From the quest of building sustainably to reflections on the future of urban design, reconciliation between a qualitative approach, which used to be dominant in project-based disciplines, and quantitative methods is actually a major challenge that must be addressed. With its subtle blend of case studies, permeated with cultural considerations and a thorough discussion of technical analysis, this publication makes an important contribution towards this reconciliation.
5
6
foreword
URBAN NE T WORK ANALYSIS
1. INTRODUCTION Design of the built environment – the spatial arrangement of buildings, blocks, streets, public spaces and the socio-economic functions they house – produces a variety of influences on urban mobility patterns and mode choices. Sprawling developments, where destinations are far apart and routes between them wide and fast, incentivize motorized trips. High density, mixed-use environments, with diverse destinations connected through a network of quality sidewalks, incentivize walking, biking and face-to-face encounter. City form and land-use patterns influence whether, how often and along which paths people choose to walk. A robust body of planning literature has emerged to articulate the qualities that make urban environments walkable and bicycle friendly. Walkability is typically associated with i) the availability of useful and diverse destinations within walking distances (e.g. retail, service and employment establishments, transit stations) , ii) safe routes that do not put pedestrians in physical or psychological danger (e.g. do not require walking next to heavy traffic, crossing wide intersections or lack barriers to separate walkways from danger zones); iii) physical and environmental comfort of the routes (e.g. step-free access, even pavements, sufficient width of sidewalks, shading from sun and rain), as well as iv) interesting routes – routes that are lined with businesses, stimulating architecture, green spaces or compelling vistas (Pushkarev and Zupan 1975; Gehl 1987; Speck 2013). However, despite a rich literature on qualities of built environments that bring people out on foot or by bike, practical methods for measuring, analyzing and modeling active mobility remain lacking in practice. Much of transportation literature on pedestrian mobility relies on rather crude proxy metrics to evaluate the walkability of a place – intersection density, block size, population or employment density and land-use mix at the census tract level are often used as predictors (Boaernet et al., 2011; Cervero and Duncan 2003; Ewing and Cervero, 2010; Hess et al., 1999; Targa and Clifton, 2005). While useful for characterizing walkability at the aggregate, whole neighborhood level, density metrics and neighborhood summary statistics do not capture the influence of built environments on mobility behavior at the individual trip scale, where decisions to undertake walks actually start.
7 Figure 1: Alley in Kampung Lawas Maspati in Surabaya, Indonesia.
8
foreword
The Urban Network Analysis (UNA) toolbox enables designers, planners and transportation scholars to measure accessibilities and predict flows of non-motorized urban movement at the individual trip resolution over networks. The software helps quantify how environmental design affects access to spatial opportunities and amenities, contributes to pedestrian flow on sidewalks, and influences the viability and patronage of amenities and public spaces in a city. These analyses not only enable us to capture the influence of urban form and land-uses on active mobility, but also inform us how planning and design decisions that shape future built environments, can be operationalized to achieve more accessible, walkable, bike-friendly and transit oriented cities. Quantitative approaches for predicting trip volumes, route choices and infrastructure utilization rates have been commonplace for motorized traffic modeling for decades. Cities use such analyses to inform transportation policy, land use policy, development rights as well as infrastructure investment decisions. The UNA tools, presented in this handbook, aim to make quantitative modeling equally accessible for pedestrian and bicycle mobility. By producing hands-on and simple-to-use software tools for modeling trips on foot or by bike, we hope to contribute to over-due efforts in rebalancing urban transportation policy from its historic biases favoring car-oriented and capital-heavy systems, towards giving more priority, specificity and quantitative rigor to urban movement that takes place on foot, by bike and other personal mobility devices (PMDs). As the name suggests, an overarching feature of UNA tools is that all spatial relationships are analyzed along networks. Whether an arrangement of rooms within a building, buildings along a street or streets within a district, the toolbox implicitly analyzes spatial relationships along circulation routes, corridors, streets or infrastructure links. Two elements of the built environment that may be close to each other along a straight-line, are not necessarily close in terms of network distance, as is the case, for instance, with buildings located on the opposite banks of a river with no bridge between them. Similarly, topological associations, such as contained or containing spaces in Euclidean geometry, do not necessarily imply access in network geometry, as exemplified by gated communities, where only limited members of society can enter. Representing spatial relationships along networks enables the UNA toolbox to describe built environments close to ways in which they are perceived by specific people or demographic groups on the ground.
URBAN NE T WORK ANALYSIS
The UNA Rhino toolbox was developed with a particular intent to make pedestrian modeling tools available to architects, designers and planners who not only investigate existing built environments, but also actively contribute to creating new ones. Most existing spatial analysis approaches are mainly used retrospectively to study existing urban developments. But the link to prescription is critical if spatial analysis is to have a meaningful effect on planning and design practice. Impact on design and planning can only be achieved if spatial analytic methods are applied in a normative way to synthetic, open-ended future design scenarios. In developing the UNA tools for Rhino – an increasingly popular and accessible software platform for designers – we have striven to incorporate measurement and analyses into a fast and iterative feedback loop, where spatial configurations can be designed, evaluated and redesigned in seamless cycles to rapidly improve the outcome. We hope the users of the tools take advantage of this functionality and not only investigate existing, but also proposed built environments. All analyses performed by the UNA toolbox require users to provide three key inputs — a Network, along which movement is analyzed, trip Origins and trip Destinations. The Origins and Destinations can optionally carry numeric attributes to weigh the analyses. Numeric attributes that indicate the number of residents in each building, for instance, can be used to weigh accessibility, footfall, or patronage estimates. Setting up a network and the appropriate Origin or Destination weights is typically the first step in any UNA application. The data needed to set up networks can either be downloaded and imported to Rhino from existing sources, such as open GIS databases, CAD base maps and Open Streetmap files, or traced directly in Rhino by the users themselves. Networks can be composed of any curve elements in Rhino (e.g. lines, polylines, arcs, splines) and form both two-dimensional and three-dimensional lattices. In preparing networks, it is important to follow the topology conventions described in Section 4.3 of this document. The UNA toolbox includes functions for importing and exporting Origin and Destination point data along with attribute information to table formats, which can be used for processing data in Excel, GIS or other applications. Having set up a network with Origins and Destinations, users can deploy a series of tools to describe and analyze pedestrian or bicycle mobility in built environments. The toolbox can be used to analyze how accessible a given set of Destinations are from a
9
10
foreword
given set of Origins along networks, which is key to understanding non-motorized trip demand to such facilities. Having determined the likely number of trips to particular destinations, users can evaluate which street segments or walking paths are likely to be utilized on such trips or estimate how many users are likely to patronize particular facilities or amenities along the way, given the distribution of demand points and competing facilities. A clustering tool allows one to detect groups of closely spaced destinations on networks, highlighting which sets of facilities might work as agglomerations, attracting more visitors. These analyses can inform what locations in a city are better or worse for particular land uses or activities, how many and what types of users public spaces or infrastructure investments are likely to benefit, or how a change in built form or land-use patterns in one location might influence pedestrian activity and facility patronage at others around it. The development of the UNA tools started as part of City Form Lab research at the Massachusetts Institute of Technology (MIT) in 2010, and subsequently moved to the Singapore University of Technology and Design (SUTD), and most recently to the Harvard Graduate School of Design (GSD). The tools are also taught and experimented with as part of three GSD courses: VIS-2129 Spatial Analysis of the Built Environment, SCI-6354 Advanced Spatial Analysis, and DES-3353 Advanced Seminar in City Form. It is a work-in-progress and intermittently updated with new functionality. This handbook is organized as follows. Part One introduces three case-studies, which demonstrate how the UNA tools have been used in practice to inform real-world urban design and transportation planning decisions. Part Two contains a technical user guide, which first covers the installation requirements and then discusses each of the UNA tools in detail. A Frequently Asked Questions section offers tips for troubleshooting. A short introductory video about the UNA toolbox, as well as a few screen-recorded tutorials can be found online at: http://cityform.gsd.harvard.edu/ projects/una-rhino-toolbox.
URBAN NE T WORK ANALYSIS
FONT CONVENTIONS
Throughout this document, different fonts are used to highlight analytic functions, variables and Rhino command line options. The following conventions are used. Italics are used to highlight UNA tool names (e.g. Betweenness), analysis input objects (e.g. Origins), as well as equation variables (e.g. beta). Machine font is used to highlight Rhino command line options available to the user (e.g. Search = 2D).
11
12
foreword
URBAN NE T WORK ANALYSIS
2. CASE STUDIES
13
14
14
URBAN NE T WORK ANALYSIS
2.1 .
PREDICTING WALKING ROUTES TO LIGHT RAIL STOPS IN SURABAYA, INDONESIA
The City Form Lab had an opportunity to collaborate with the World Bank and an Australian urban design firm — Hansen Partnership — on developing a transit oriented strategic plan for downtown Surabaya in 2013-14. In the course of this work, UNA tools were extensively used to predict future walking activity around proposed tram stops and to make informed suggestions as to which access streets around future tram stations should be prioritized for infrastructure upgrades, urban design and landscaping improvements, so as to ensure that the future light-rail ridership is maximized, particularly among women and children, who were expected to constitute most important groups of future transit users. The following case study describes the application of UNA tools to address Surabaya’s TOD planning challenges (City Form Lab and Hansen 2015). 2.1.1. THE MRT PROJECT
With a current municipal population of 3.1 million and a metro area population of 5.6 million, Surabaya is the second largest city in Indonesia after Jakarta. Facing a rapidly growing population, expanding economy and worsening traffic congestion, the City of Surabaya developed a bold proposal to implement a light rail system. At the time, it was not only going be the first rail-based urban mass transit system in Surabaya, but the whole of Indonesia. The Surabaya Integrated Mass Rapid Transit System (SMART), involved two key components — a north-south street level tramway (SuroTram) and an east-west elevated monorail (BayaRail). A pre-feasibility study identified the north-south light-rail corridor as a first priority undertaking, leaving the east-west monorail as a
15 Figure 2: Kampung in Surabaya, Indonesia
16
surabaya , indonesia
second-order priority for the future. The proposed tram corridor (Figure 3) follows the historic north-south spine of the city center, starting near the southern crossing of Wonokromo and Kalimas rivers and ending close to the port in the north. Along the way, the tram passes through historic districts, markets, commercial areas, kampungs (traditional urban villages), and the CBD commercial district. Most of the city’s existing trips in 2013 occurred on motorcycles and scooters, whose presence all over South East Asia rapidly grew since the late 1990s. Cars made up around seven percent of all trips and privately operated and ticketed minibuses around five percent. Thirty four percent of all trips were non-motorized — on foot, by bike or rickshaw. The light-rail proposal presented a unique opportunity for the city to maintain and bolster non-motorized and light-motorized travel, before private vehicle ownership starts dominating. Both the light-rail system and related transport oriented development initiatives around the rail stood to play a key role in capturing this opportunity. Trams are fundamentally pedestrian oriented transportation systems — a vast majority of tram riders walk to the tram and from the tram to their destinations. Through good infrastructure provision and deliberate urban design investments, it is also possible to attract riders to walk or bike to tram stops from considerable distances. The project team found it unlikely, however, that initial ridership would come from vehicle users, who would drive to the tram, park away their vehicle, take the tram to their destination, walk to their destination from there and repeat the same procedure on the way back. Scooters offer door-to-door convenience at a lower cost than the tram, which the newly proposed system was unlikely to rival. Cars additionally carry a status symbol in Indonesia. We thus stipulated that in order to attract ridership to the tram, provisions have to be made to integrate the tram stops carefully with the surrounding Origins and Destinations on foot and by bicycle. We also stipulated that the most likely riders for the tram system will be middle and lower-middle income citizens, many of whom live in kampungs around the proposed route. Lower and middle income women were seen as particularly likely riders, given their lack of access to private vehicles. Attracting pedestrians and bicyclists to walk and bike to tram stops necessitates a comfortable and convenient network of pedestrian pathways and safe bike routes leading to the stations. Many
URBAN NE T WORK ANALYSIS
17 Figure 3: 800m buffer corridor along the tram line. Key Tram stations
Tram line
Streets
( !
Water
( !
( ! ( ! ( ! ( ! ( ! ! ( (! ! ( ! ( ( ! ( ! ! ( ( ! ( ! ( ! ( (! ! ( ! ( ! ( ! ! ( ( ! ( !
( ! ( ! ! (
( ! ! (
18
surabaya , indonesia
streets in Surabaya are in poor condition for pedestrians. Sidewalks, if they exist, are discontinuous, lack ramps at intersections, remain poorly lit at night and are sometimes unpaved or contain potholes and other obstacles. Many city center streets are routinely flooded when it rains, due to a clogged or otherwise disconnected drainage system. During the past two decades, urban infrastructure investments in Surabaya have primarily prioritized motorized traffic — wider lanes, higher vehicle flow, pedestrian barricades to prevent jaywalking, footbridges over streets to separate traffic. These interventions have inadvertently discouraged pedestrian activity on streets and curbed the development of pedestrian-oriented land uses, such as street-facing shops, restaurants and services in the city center. Significant planning, investment and management was needed to make streets walkable and pedestrian friendly. And since undertaking such investments throughout the entire city center was practically unaffordable, it was essential to identify priority streets that would most directly affect tram ridership and where investments should flow first. The analysis presented below illustrates how such streets were identified for subsequent on-site evaluation and potential urban design improvement proposals. 2.1 . 2. PEDESTRIAN
NETWORK ANALYSIS AROUND THE TRAM
Three different analyses were performed within an 800-meter buffer (10-15 minutes’ walk) around the proposed tram line: population density analysis; tram catchment analysis around each stop; and the identification of key pedestrian feeder streets. 2.1 . 2.1 .
Population Density
Population estimates of the census blocks (called RTs in Indonesia) surrounding the tram corridor were assembled, converted into density estimates and visualized along the tram corridor (Figure 4). The data showed that approximately 263,000 residents lived within an 800m walking radius around the proposed tram stops. This result was determined with the Service Area tool in the UNA toolbox. The Service Area tool selects all Destination points that can be reached within a specified travel radius around a set of Origin points. The Origin points in this case were tram stops and the Destination points were census block centroids, using their
URBAN NE T WORK ANALYSIS
19 Figure 4: Population density in census blocks (RTs) along the tram line
P !
Key   Tram stations
P !
Tram line 5615
P !
1052
0
Population Density (per hectare)
P ! P ! P !
P ! P ! P ! P !
! P P !
P ! ! P P !
20
surabaya , indonesia
population numbers as weights. An 800m Search Radius was applied to ensure that only those census blocks that were within an 800m walk from at least one of the stations were included. Since the UNA toolbox operates on street networks, the specified 800m walkshed was analyzed along street networks instead of simplified straight-line distances. Having selected all the blocks that were within an 800m walkshed around stations, the “attribute table” of these points was exported to Excel, using the Export Data tool, where the population weights were summed. The overall catchment estimate of 263,000 residents found within the 800m walkshed around the tram corridor could be slightly inflated, since it also includes census blocks around the split tram alignment, where the combined distance from the nearest tram stop in each direction is actually longer than 800m due to the extra distance between departure and return stations. The population density varies widely along the corridor, with an average density of 471 and a maximum of 5,615 people per hectare. 2.1 . 2. 2.
Catchment areas of tram stations
Estimated population catchment of each station was computed with Accessibility Indices using Reach analysis, which summarized the population numbers that could be reached within the given walk radius around each station. Reach analysis was applied three times, each time with a different radius: 200m, 400m, and 800m (Figure 5). The results show that 14,733 residents lived within 0-200 meters (0-3 minutes’ walk) from tram stations. Another 53,931 people within 200-400 meters (3-7 minutes’ walk), and 194,471 people within 400-800 meters (7-14 minutes’ walk). Estimated ridership catchment within the total 800m walkshed varied widely between stations – ranging from over 20,000 people around some stations to less than 3,000 around others. We learned that the largest share of potential riders would need to walk 7 to 14 minutes to reach their nearest tram stop and the same distance back home from tram stops. This is a substantial amount of walking according to international transit precedents. In order to make sure these trips happen, walking paths between homes and tram stops need to be safe, comfortable and attractive. These walking routes should not flood during heavy rain; they should be well paved and suited for strollers and carts, lit for access after dark, landscaped with greenery and desirably lined with activity generating uses, such as retail, service and rec-
URBAN NE T WORK ANALYSIS
21
P !
P !
A: 427 B: 1,527 C: 12,020
Figure 5: Catchment of tram stations, Reach Analysis and Service Area Analysis
KeyKey
A: 786 B: 2,964 C: 21,622
P ! A:753 B: 1,777 C: 15,326
P ! P !
A: 354 B: 3,601 C: 14,866
A: 1,603 B: 3,996 C: 19,450
P !
A: 496 B: 2,469 C: 14,605
P !
A: 775 B: 3,534 C: 17,203
A: 1,240 B: 2,236 C: 19,133
P ! A: 1,901 B: 5,304 C: 19,123
Catchment
P ! P !
P !
A: 683 B:1,481 C: 8,784
! P
A: 689 B: 3,297 C: 15,297
A: 1,141 B: 3,356 C: 17,170
P Tram Tramstations stations Tram line Tram line Catchment 0-200m 0–200m (A) 200–400m 200-400m (B ) 400-800m (C) 400–800m
A: 301 B: 1,879 C: 12,016
P ! A: 257 B: 1,669 C: 19,316
! P P !
A: 337 B: 1,490 C: 13,361
22
surabaya , indonesia
Figure 6: Average footfall to tram stations, Betweenness Analysis
P !
Key  Key Tram stations P ! Tram stations Tram line Tram line 16762 16 762
P !
Footfall
500to stations 2700
0 1001
P !
(per day)
P !
Footfall to stations (per day)
P ! P !
P ! P ! P ! P !
P ! ! P
! P ! P P !
URBAN NE T WORK ANALYSIS
23 Figure 7: Average footfall to tram stations in the whole tram corridor, Betweenness Analysis ! !
Key Key Tram stations P ! Tram stations Tram line Tram line 16762 16 762 ! !
! !
Footfall
500to stations 2700
0 1001
(per day)
Footfall to stations (per day)
! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !
! ! ! ! ! ! ! !
! !
! !
! !
! ! ! !
! ! ! !
24
surabaya , indonesia
reational amenities to facilitate pedestrian use. The project team recommended that the city consider an upgrading program to evaluate the current conditions of these critical streets and implement necessary improvements to maximize pedestrian accessibility to the tram. 2.1 . 3 . FEEDER
STREETS
The extent of the street network that connects the tram stations to all the census blocks within a 800m walking radius spans a cumulative length of 269 km or 167 miles (Figure 7). The Service Area tool in the UNA toolbox can be used to select Destinations or street segments within a specified radius around Origin points, and the distances of the selected segments can be summed to arrive at the above result. It would be a colossal task to upgrade all of these streets. In order to assess which streets are most critical for serving pedestrian access to the tram, and in order to prioritize potential upgrading efforts, we performed the following detailed analysis of feeder streets. We estimated the plausible walking paths from each census block to the nearest tram station and routed the equivalent of each census block’s population to the corresponding walk to the station. Repeating the same procedure on all of the 1,001 census blocks that fell within an 800m walking reach of the stations produced a cumulative pedestrian footfall estimate on each street segment around the tram corridor (Figures 6 and 7). This analysis was achieved with the Betweenness tool in the UNA toolbox, which calculates the Betweenness value of each street segments along routes from each census block to its nearest tram station, weighted by the population of each census block. Trips are cumulatively added, which means the more used a street segment is, the higher Betweenness value it gets. Since the Betweenness index is weighted by population numbers, each street segment gets a value, which represents the number of estimated pedestrians crossing it. This data is then documented and visualized spatially with a color gradient as seen in Figures 6 and 7. A Detour Ratio input in the Betweenness tool allows the user to model trips that are longer than the shortest routes by a specified percentage. Pedestrian route choice surveys have found that people commonly deviate 15-20% longer than shortest routes, which suggested we use a Detour Ratio value of 1.15 or 1.2, more
URBAN NE T WORK ANALYSIS
accurately representing the modeled behavior and route choice. Additionally, a distance decay function in the Betweenness tool enables the analysis to gradually diminish the number of trips routed to destinations as walking distances increase, capturing the effect of travel costs. Since the estimates show the total number of residents that would rely on a given street segment on walks to their nearest tram station, it would be reasonable to prioritize upgrading works on routes that are most utilized. Upgrading works should ideally be continuous, without leaving problematic path segments between upgraded sections. The footfall analysis showed that a number of key walking paths to the proposed tram stations pass through Kampungs and involve both larger public sidewalks as well as small kampung lanes. The highest population density along the tram corridor was found near Joyoboyo, Keputran, Tunjungan, Genteng and Kembangan Buyut, where residential densities of over a thousand people per hectare are found. Priority should be given to path upgrades in these areas. In Figures 8 and 9, the most trafficked path segments along the whole corridor (those with flow estimates over 1000 people/day) were filtered out, providing a selection of the most critical paths to the future tram stops. These segments cumulatively form 33 kilometers of pathways that could be assessed for pedestrian quality, and where necessary, improved accordingly. The City Form Lab and Hansen study (2015) put forth a series of upgrading initiatives on these critical feeder routes. 2.1 . 4 . DISCUSSION
The above analyses of pedestrian accessibility to the tram stations only accounted for residential access. We were unfortunately not able to assemble data on the distribution of employment locations and business establishments in Surabaya. A more holistic evaluation of all types of tram ridership walks — coming from or going to homes, workplaces, retail, service or recreation destinations — would be preferable, where such data is available. The methodology presented in this case-study is equally valid for cities whose street infrastructure is in a different condition than Surabaya’s. In Surabaya, the detection of most critical walk routes to public transit stops would enable a city to fix drainage around them, so they don’t flood when it rains; to pave proper sidewalks;
25
26
surabaya , indonesia
Figure 8: Street segments with over 1000 average footfall per day, Betweenness Analysis
P !
Key  Key Tram stations P ! Tram stations Tram line Tram line 16762 16 762
P !
Footfall
to stations 2700 2700
1001 1001
P !
(per day)
! P
Footfall to stations (per day)
P ! P !
P ! P ! P ! P !
P ! ! P
! P ! P ! P
URBAN NE T WORK ANALYSIS
27 Figure 9: Street segments with over 1000 average footfall per day in the whole tram corridor, Betweenness Analysis
( !
Key Key Tram stations P ! Tram stations Tram line Tram line 16762 16 762
( !
( !
Footfall 0 to stations (per day)
1
to stations 2700 2700
1001 1001
(per day)
Footfall to stations (per day)
( ! ( ! ( ! ( ! ! ( (! ! ( ! ( ( ! ( ! ! ( ( ! ( ! ( ! ( (! ! ( ! ( ! ( ! ! ( ( ! ( !
Tram stations Tram line
62
Footfall
( ! ( ! ! (
! ( ( !
28
surabaya , indonesia
to install street lighting and furniture; and to zone the adjacent properties for activities generating pedestrian uses, such as retail and service activities. In different cities, a prediction of key access routes to transit stops could lead to changes in zoning codes or building design guidelines along such routes, to reductions in motorized traffic on surrounding roads or to more extensive landscaping or urban design investments. In snowy climates, an analogous study could identify highest priority sidewalks for municipal snow plowing. Knowing where to prioritize public money is always important. In order to validate the accuracy of footfall predictions produced with the above Betweenness algorithm, empirical pedestrian surveys can be carried out. For instance, Sevtsuk (2018) illustrates how a similar analysis was used to predict peak hour footfall around a subway station in Cambridge, MA. Predicted pedestrian volumes on specific street segments explained around 85% of the empirically counted pedestrian volumes on the same streets. For more about the Surabaya project, please see http:// cityform.gsd.harvard.edu/projects/surabaya-mrt-corridor-concept-plan
30
surabaya , indonesia
URBAN NE T WORK ANALYSIS
2. 2.
PLANNING RETAIL CENTERS IN PUNGGOL, SINGAPORE
The City Form Lab worked with the Singapore Housing and Development Board (HDB) on a strategy for commercial facilities in Punggol, one of HDB’s newest public housing towns. Among other tasks, the project required locating and right-sizing retail centers for the new town. UNA tools were extensively employed to predict the patronage of planned retail centers and to simulate best locations and sizes of retail centers. This case study illustrates how UNA tools can be used to inform spatial facility location and sizing. It focuses on retail clusters, however, the same tools can be applied to other types of planned facilities, such as city parks, playgrounds, shared bicycle stations, public libraries etc. 2. 2.1 . PUNGGOL:
A TRANSIT-ORIENTED RESIDENTIAL TOWN
Punggol is one of the newest residential towns in Singapore, projected to house around 300,000 residents when complete. It is a transit-oriented residential new town, connected to the city center by a Mass Rapid Transit (MRT) line from the Punggol MRT station. The town’s sub-districts and precincts are connected to the MRT trunk via an elevated Light Rail Transit (LRT) system, local buses and walkways. Commercial center planning in HDB towns typically follows a hierarchical structure of town, neighborhood and precinct centers. Punggol is envisioned to have one town center, seven neighborhood centers, and around 29 precinct centers. At present, Singapore’s HDB towns have 9% of total retail space in town centers, 44% in neighborhood centers and 47% in precinct centers (MTI Economic Review Committee 2002). Commercial amenities are essential to vibrant urban neighborhoods. Having shops, restaurants and personal service es-
31 Figure 10: An LRT station and bus stop in Punggol, Singapore.
32
punggol, singapore
tablishments near places of residence or employment not only increases people’s choices but also encourages walking, reduces urban energy usage, fosters social cohesion and generates local jobs. It was therefore important to ensure that Punggol’s retail centers constitute a robust district-wide commercial system that maximizes access to residents. In order to accomplish this, UNA tools were used to predict patronage of planned retail centers and to simulate better locations and sizes of retail centers that maximize patronage. 2. 2. 2. THE
HUFF MODEL AND MODIFICATIONS
UNA facility patronage analysis is based on the widely used retail expenditure model developed by David Huff (1963). The Huff model postulates that visits to each retail center are proportional to the attractiveness of the destination and inversely proportional to its distance or travel time from patrons. However, the implementation of the Huff model in the UNA toolbox includes important modifications that enable the Huff model to work on spatial networks and allow users to control transportation cost variables and decay rates in ways that link total visits in a system to its spatial configuration (Sevtsuk and Kalvo 2017). 2. 2. 3 . PATRONAGE
ANALYSIS
The study set out to predict the patronage of commercial centers that were already built or planned by HDB, as well as to figure out possible improvements to the spatial pattern of these centers in ways that would maximize collective patronage to all retail destinations in town through future developments. Figure 11 shows the town layout, including the street network, building footprints, and the locations of existing, built-out commercial clusters, with their approximate size indicated in square meters. The total housing stock, which forms the Origins of retail trips, consists of 96,112 dwelling units. Figure 12 shows the predicted patronage of each of the centers according to the traditional Huff model, using of the Find Patronage tool in the UNA Rhino toolbox. This tool estimates the patronage of facilities located on a network based on a given distribution of demand points and competing facilities. Although building centroids in Figure 12 are shown as uniform blue dots, each of them carries a different weight, W, indicating the number
URBAN NE T WORK ANALYSIS
33 Figure 11: Punggol town layout with its street network, building footprints and existing commercial centers. Numbers indicate the Net Leasable Area of each existing commercial cluster.
151
Key
907 1176
Neighborhood Center
804 111
801
55
1015
802
Precinct Center Building footprints
864 8050
Figure 12: Estimated store patronage with existing stores in Punggol. Total patronage in town = 96,112 households. Beta = 0.001; search radius = 3000 meters; alpha = 0.37. Key
5477 9862 4306
3412
10540 11879`
9083
Neighborhood
Key Center
8784
8999
Neighborhood Center Precinct Center Precinct Center Points BBuilding uilding Points
7831 15935
34
punggol, singapore
of dwelling units at each building. These demand points weighted by the number of dwelling units are used as Origin points in the analysis while retail centers, weighted by their sizes, serve as Destinations. Individual Destination results in Figure 12 suggest that estimated daily patronage varies from 3,412 households at the smallest center to 15,935 at the largest. Overall patronage across all retail centers is 96,112 — the same number as the number of households — as the Huff model requires. The traditional Huff model assumes that all patronage or purchasing power is fully spent among available destinations. Because of this assumption, overall patronage across all stores is not affected by different environmental configurations – even if all stores were concentrated in single, large cluster in the center of town, the model would still estimate 96,112 visits to that center. This is not the case in reality because transportation costs affect store patronage when demand is elastic. Households with less access to retail centers tend to visit them less frequently (Sevtsuk and Kalvo, 2017). In order to account for unequal access to destinations, an additional distance decay function was incorporated in the Find Patronage tool, which can be triggered by the ApplyImpedance user option. Patrons' sensitivity or elasticity to distance is controlled by a distance decay coefficient beta, which was empirically determined as "0.001" (distance units in meters) for HDB town retail trips. In Figure 13, store patronage is estimated on the same configuration as in Figure 12, but applying the additional “distance decay function”. The relative number of visits to each center remains similar, but the additional “decay effect” drops overall patronage by 65%, reducing total patronage from 96,112 (the number of total households in the area) to 33,211, as one would expect. The town layout, and in particular the distance needed to get to stores, now affects overall patronage — when stores are more accessible to residents, patronage increases. 2. 2. 4 . CONSIDERING
PUNGGOL’S TRANSIT-ORIENTED CHARACTERISTICS
In both estimations shown in Figures 12 and 13, shoppers are assumed to start their trips from residential buildings. However, shopping from home does not necessarily reflect the dominant pattern of store visits. In Singapore’s HDB towns, over 65% of
URBAN NE T WORK ANALYSIS
35 Figure 13: Estimated store patronage with existing stores in Punggol using a distance decay effect. Total patronage in town = 33,211 households. Beta = 0.001; search radius = 3000 meters; alpha = 0.37.
2157 3355 1525
1092
3792
3193
Key Neighborhood
2865
2721
3593
2661 6252
residents use public transit, including MRT, LRT (Light Rail), and bus, for daily home–work–home commutes. Locating retail centers conveniently en route to transit stops could enable residents to patronize stores without incurring extra transportation costs that accrue with designated trips from homes. In order to model retail demand on routine walks between homes and transit stops, the “demand weights” found at home locations can be distributed along walkways leading to public transit stops at given distance intervals (e.g. 10 meters) using the Distribute Weights tool in the UNA toolbox. If a building’s original demand weight was “100 dwelling units,” for instance, it was located 1,000 meters from the nearest public transit stop, and the distributed demand points are placed at 10-meter intervals along the walk, we obtain 100 “distributed demand points,” each getting “1” as their demand weight. And when multiple routes overlap on particular segments of the circulation network, then the same distributed points are used and their values are cumulatively summed. As a result, points on highly trafficked street segments, such as those near MRT, LRT and bus stops, accumulate higher weights. The Distribute Weights tool allows us to re-distribute point weights from stationary Origin locations to network routes
Key Center
Neighborhood Center Precinct Center Precinct Center Points BBuilding uilding Points
36
punggol, singapore
that lead to chosen Destinations. The choice of routes in the tool relies on the Betweenness algorithm previously described in the Surabaya case study. As in the Betweenness tool, the Distribute Weights tool does not necessarily assume that pedestrians use the shortest route. Using a DetourRatio variable on the command line enables the tool to choose routes that are a certain percentage longer than the shortest one. Based on a survey of pedestrian activity in HDB towns, we used a 15% DetourRatio to allocate household weights evenly between all routes that are up to 15% longer than the shortest path from homes to transit stops. A conditional decision tree, shown in Table 1 below, was used to determine how much of each household’s demand weight was allocated to routes leading to each type of nearest transit stop, depending on stop availability within an 800m walkshed. As a result of distributing the original demand weights from buildings to walking routes to transit stops, the sum total of the weights does not change — it stays at 96,112 corresponding to the number of original households in the area. Table 1: Conditional decision tree used to determine how much of each household’s demand weight in Punggol is allocated with the Distribute Weights tool to routes leading to each type of nearest transit stop, depending which stops were available and how far around each residence.
Present within 800m network radius Bus
LRT
MRT
If
TRUE
TRUE
TRUE
If
TRUE
TRUE
If
TRUE
If
TRUE
If If If If
Bus
LRT
MRT
Then
10%
30%
60%
Then
30%
0%
70%
Then
30%
70%
0%
Then
100%
0%
0%
TRUE
Then
0%
30%
70%
TRUE
Then
0%
0%
100%
Then
0%
100%
0%
Then
0%
0%
0%
TRUE TRUE TRUE
Weight
Figure 14 describes the “distributed demand weights” on estimated walking routes from each dwelling unit to the nearest MRT, LRT and bus stop. On top of these, the figure also shows the new patronage estimates for the same set of existing stores as we saw in Figure 13. Overall, patronage across all stores slightly increases from the previous 33,211 to 35,055 as a result of modeling demand from walks to transit stops rather than homes (5.5% increase). This suggests that placing future retail developments along popular walking routes to transit stations would increase retail accessibility for Punggol's residents.
URBAN NE T WORK ANALYSIS
37 Figure 14: Estimated patronage of existing clusters with demand originating from walk routes to transit stops.
2311 4008 1398
Total patronage in town = 35,055 households. Beta = 0.001; search radius = 3000 meters; alpha = 0.37.
3428
3724
Key 1106 2909
2708
3973
2824 6661
2. 2. 5 . SIMULATING
NEW STORES
At the time of the study, Punggol was roughly half built, with less than half of the planned commercial spaces constructed. A town center — the largest commercial center in the plan, located at the Punggol MRT junction — had not yet broken ground. A number of neighborhood centers were also in planning phase. A few precinct centers were complete. The partially built nature of the project offered an opportunity to explore how the remaining commercial space could be best positioned and sized so as to maximize overall retail access in the town. Two planning scenarios with different location patterns were compared using the same distributed points on MRT walking routes shown in Figure 14 as demand locations. The first scenario reflected a pre-existing plan for the distribution of future commercial centers in Punggol, including both those that were already built out as well as those that remained yet to be built. The second scenario illustrated a tabula rasa approach, where the same number of centers was positioned at optimized locations to maximize access from MRT and bus stop walking routes. The following analyses
Neighborhood
Key Center
Neighborhood Center Precinct Center Precinct Center Distributed Distributed Demand Weights
Demand W
38
punggol, singapore
examine how the positioning and sizing of stores impacted overall retail patronage in both scenarios. The two scenarios were made comparable by making sure that both contained one town center (30,000m² net leasable area), seven neighborhood centers (9,000m² each) and 29 precinct centers (1,500m² each). The overall quantum of retail space was kept constant, corresponding to the amount foreseen for Punggol. Figure 15 illustrates the results for scenario one, where existing and future commercial clusters are located according to pre-existing plans for Punggol. Total patronage across all commercial centers is estimated at 38,243 households. This result increases to 38,899 in scenario two (Figure 16), when the same number and size combination is shifted closer to the most trafficked MRT walkways. Although the improvement resulting from slightly closer store locations is small (1.7% increase), it could be more substantial in other contexts that have less stores and bigger market areas. Another way to affect patronage at a given set of Destination centers is to reallocate Destination sizes. This is because Gravity accessibility outcomes simultaneously depend on both Destination size (attractiveness) and distance (see Section 4.4.1.2.). In Figure 16, we optimized retail Destinations to be closer to MRT walking routes, but used a typical size allocation of HDB centers — 30,000m² for a town center, 9,000m² for neighborhood centers and 1,500m² for precinct centers. Keeping total retail space constant, we explored whether overall patronage could be increased with a different center-size patter. The unaPatronageSim tool, which has no graphic user interface button and can only triggered from the command line, is a simulation tool that tests what size combination of retail centers achieves most estimated visits. Similar to above, inputs include Origin points, determining where retail demand is coming from and retail Destination locations. Additionally, a total retail area limit to be allocated between centers is required. This area is then iteratively allocated between all center types at a chosen percent interval — 1% at town center and the rest between other types of centers, 2% at town center and so on, until 100% has been tested at each center type. Just like above, the attraction of each retail cluster depends on its size and network distance from potential customers, and the highest result is obtained when accessibility to stores is maximized with respect to a given demand pattern. This accessibility, in turn, depends both on the proximity and size of destinations, and the beta and alpha coefficients, which describe
URBAN NE T WORK ANALYSIS
39 Figure 15: Patronage estimate for scenario one, according to the pre-existing plan. Total quantum of commercial space is 136,500m².
735 734 1182
500 570
732
785 567
849 820
932
680
814
3209
910
949
953
799 750
1973 975 1062
951 1879
902
973
904
649
877
Key
969
1149
1040
Town Center Neighborhood Center
2367 2234
762
544
583
Precinct Center
Figure 16: Patronage estimate for scenario two. Total quantum of commercial space is 136,500m².
544 1710 542
608 478
Estimated patronage across all clusters = 38,899 households. Beta = 0.001; search radius = 3000 meters; alpha = 0.37.
812 1601 595
758 1692 618 519
1052
3510
818
928
635
1849
986 939 795 834
848
990 823
1952 1003 736
Total patronage in town = 38,243 households. Beta = 0.001; search radius = 3000 meters; alpha = 0.37.
544
Town Center
1127 2109 1080 1098
2128
555
Key
522
Neighborhood Center Precinct Center
40
punggol, singapore
people’s sensitivity to travel costs and Destination size respectively. Our use of “0.001” for beta and “0.37” for alpha was determined from an empirical survey of household shopping trips across HDB towns in Singapore. Figure 17 illustrates simulation results, where a given quantum of total net leasable area (NLA) was iteratively allocated at 5% intervals to different center types. The graph depicts the resulting patronage estimates on the vertical axis and the percent of total space given to the town center on the horizontal axis. The sum total of retail NLA is kept constant in each iteration, while different amounts of retail space were allocated between different center types. Within each zigzag hump in the graph, allocations move from 100% at the precinct centers on the left, to the reverse, 100% at neighborhood centers on the right. The overall pattern in the graph suggests that with every town center proportion, patronage is highest when all of the remaining retail space is only allocated to medium-sized neighborhood centers and none to the smallest precinct centers. The maximum result of 41,254 is achieved when the town center and seven neighborhood centers are all the same size — 17,065m². This result suggests that a network of bigger neighborhood centers could benefit both residents and businesses more than the current hierarchical structure with a single large town center, a few medium-sized neighborhood centers and lots of small precinct centers. In other words, Punggol's retail patronage would benefit from having no small precinct centers at all, instead boasting more mid-size neighborhood centers that produce a bigger customer draw. Note that the simulation results do not have a single sharp peak — there are multiple peaks with town center sizes ranging from 5 to 55%, where overall total patronage remains within 1% of the maximum. Even with a 75,500m² town center (55% of total retail space in town) and seven 8,775m² neighborhood centers, patronage reaches 40,896 households. No one center size configuration is clearly above others, but overall patronage is maximized when retail floor area is distributed among seven or eight (if counting the town center) neighborhood centers, each accommodating around 9,000–18,000m² of space. These results apply to Punggol. A different size combination may work in other towns, depending on their layouts and customer densities. However, since the most important determinants of optimal size configuration — the alpha and beta values — were based on survey data that was combined from nine HDB towns through-
URBAN NE T WORK ANALYSIS
41 Figure 17: Patronage simulation results, where the total quantum of 136,500m² was iteratively allocated to town, neighborhood and precinct types using allocation 5% steps. Maximum patronage is 41,254 households. Beta=0.001; search radius=3000 meters; alpha=0.37.
out Singapore, a stronger emphasis on medium-sized neighborhood centers should also be explored in other HDB towns. Overall, optimizing the sizes of retail clusters within the same locations, as in scenario two (Figure 16), increased estimated retail patronage from 38,899 to 41,254 (6%). Destination size optimization thus had a bigger effect on patronage than location adjustments above. But if we take the location optimization that placed centers closer to MRT walking routes and size optimization together, the combined gain compared to the baseline scenario was a 10.2% increase in estimated retail visits — a considerable improvement for the residents of Punggol and shop owners alike. 2. 2.6 . OTHER
APPLICATIONS
The model and relevant UNA tools described enable planners and urban designers to better understand how spatial planning could influence the viability of retail developments. Successful planning of urban commerce requires that stores receive sufficient customers and revenue to break even. The siting of buildings, their densities and the layout of circulation networks play an important role in shaping the demand that sustains urban retailers. In the Punggol case, we focused on optimizing retail locations and sizes. But similar gains could be achieved by manipulating demand locations and access routes – housing site plans, residential and employment densities, pedestrian paths as well as transit locations. If applied in an existing retail system in other cities, the model could be used to assess how an expansion of a given cluster of stores could impact patronage at the named, as well as competing clusters around it. This could inform policy makers on
42
punggol, singapore
how and where zoning alterations and policy incentives are best placed to bolster local economic activity. An analogous model can be specified for different types of urban facilities, using appropriate coefficients and parameters for each. The planning of urban parks or playgrounds, for instance, could benefit from a patronage model that too aims to maximize overall usage. The placement of electric car chargers and shared bicycle stations face similar issues. UNA Find Patronage and unaÂPatronageSim tools are designed for a wide range of spatial facility patronage applications.
44
punggol , singapore
URBAN NE T WORK ANALYSIS
2. 3 .
HOW ACCESSIBILITY EXPLAINS RETAIL AND FOOD ESTABLISHMENTS' LOCATION PATTERNS IN CAMBRIDGE AND SOMERVILLE, MA
This case study analyzes retail and food establishments' location patterns in Cambridge and Somerville, MA in order to understand which spatial factors explain the observed store pattern. The study uses approximately fourteen thousand buildings as units of analysis, testing how pedestrian accessibility from surrounding buildings, land uses, and transit stations affects the probability of observing retail and food-service establishments in individual buildings. The Gravity index in the Accessibility Indices tool and the Betweenness tool were extensively used to measure the location and accessibility qualities around all stores. 2. 3 .1 . STUDY
AREA
Cambridge and Somerville, MA, located across the Charles River from downtown Boston, are spatially continuous and similar in size. The land area of Cambridge is 6.43 square miles; it housed 101,388 inhabitants in 2007. The land area of Somerville is 4.1 square miles; it housed 74,405 people in 2007. With average population densities of 25 inhabitants per acre (62.5 people/ hectare), Cambridge and Somerville illustrate moderately dense east-coast urban environments. Their richness in pedestrian and transit commuters and their well-known patterns of retail clusters around a number of “squares” made them particularly attractive for this study.
45 Figure 18: Harvard Square commercial cluster, Cambridge, MA
46
cambridge and somerville, ma
Data from 2009 ESRI Business Analyst included a total of 1,941 individual establishments in Cambridge and Somerville: 1,258 retail establishments (NAICS 44-45) and 683 eating or drinking establishments (NAICS 722). Geographic coordinates, as well as an address field associated with each establishment enabled matching each business to a particular building in the two cities. Each building thus obtained a binary dependent variable (0 or 1) showing whether or not it contained retail or food establishments. This binary variable became the dependent variable of a regression model that analyzed the probabilities of observing retail or food service establishments across both towns. The original dataset had 26,983 individual buildings in Cambridge and Somerville, out of which 961 contained one or more retail or food establishments. Not all of these buildings, however, were suitable for a retail and food establishment location choice model. Zoning allowed commercial uses in only certain areas of both cities and permitting the model to include all buildings as part of a location choice universe was deemed to be unrealistic. We eliminated all zoning blocks that were designated as “single family” housing from the data and kept only those buildings that were commercial, industrial, multifamily, or mixed as a choice set. However, we then found that a number of businesses were currently located within “single family” zones. This could be the result of “planning variances” and “special permits” in the area. To account for such conditions, we drew a 100-meter buffer around each of the businesses in residential areas and re-included the buildings within these buffers to be part of our retail location choice set. The final building count that was included dropped from 26,983 to 14,218, or 52 percent of the original buildings. As shown in Figure 19, 834 of these buildings contained a retail or food service establishment, some of which can be seen clustered around Harvard Square, Central Square, Inman Square, and Union Square. Employment estimates were obtained from the ESRI Business Analyst data, which indicates the estimated number of employees per each establishment. Residential estimates came from the block-level census 2000 data, transit networks from MassGIS, parcel characteristics from the two cities’ assessor databases, and road and sidewalk characteristics from Tiger roads data. Since census counts did not originally come at an individual building level, they had to be interpolated between residential buildings. To do so, we first filtered out only residential buildings
URBAN NE T WORK ANALYSIS
47
in each census block according to assessors’ data and then allocated the census counts proportionately between the remaining residential buildings, weighing the allocations by building volume. The total sum of residents remained identical to census estimates in each census block. Figure 19: Observed locations of retail and food- service establishments in Cambridge and Somerville, Massachusetts. Key Buildings (n = 26,983) Buildings used for analysis (n = 14,218) Buildings containing retail and food establisments (n = 834)
The above geographic data was used to build the network for spatial analysis in Rhino. Building entrances were represented with points, positioned at the centroids of actual building footprints, and we assumed that each entrance connected to a street that lies closest to it along the shortest perpendicular connection. Each building point had attributes to describe it: estimates of resident counts, employment counts, built volume in GFA (gross floor area), whether or not it is next to a transit stop, as well as whether or not it contained a retailer (0 or 1, as binary variables). The analysis focused on one mode of travel: walking. We tested how pedestrian access from different Origins — homes, workplaces and transit stations — affected the likelihood of observing retail and food businesses in buildings across Cambridge and Somerville. These variables did not describe actual, observed walking trips per se, but rather the potential accessibility to the different Destinations around each building in a 600-meter network radius (approximately 10-minute walk). The analysis not only tested how retailers locate with respect to pedestrians at trip Origins — homes, workplaces, or comple-
The extent of gray buildings marks the municipal boundaries of the cities.
48
cambridge and somerville, ma
mentary businesses — but also with respect to bus and subway stations, which bring a notable portion of shoppers to stores, especially in larger clusters like Harvard Square. The study did not estimate how access by car affects retail patterns in the area — this is one of its limitations. Observations of commerce in Cambridge and Somerville suggested, however, that most trips to stores arrive on foot or by transit. 2. 3 . 2. INDEPENDENT
VARIABLES: ACCESSIBILITY MEASURES
2. 3 . 2.1 .
Gravity Analysis
The Gravity accessibility index (Hansen 1959) measures the ease with which a set of Destinations can be accessed from a set of Origins. Here, we illustrate three applications of Gravity analyses performed using the UNA Accessibility Indices tool, quantifying access form the same set of potential retail sites (or 14,218 Origin buildings) in Cambridge and Somerville to three different Destination types: surrounding residents, built volume (GFA), and subway stops, all within a 10-minute walking radius (r = 600m). These accessibility results were subsequently analyzed to determine which location qualities retail and food service establishments are statistically drawn to. First, in order to capture spatial variations in residential density, we measured accessibility from all origins to surrounding residents. All buildings within the boundaries of Cambridge and Somerville plus some additional buildings in a 600m buffer around both towns were selected as Destinations. Adding a buffer of Destinations around the perimeter of the study area, with a similar 600m radius as specified in the accessibility analysis, is a common technique for avoiding an “edge effect” that can otherwise result from artificially truncating spatial data. The numbers of residents in each Destination building were set as Destination weights in the analysis. Destinations with no residents consequently did not affect accessibility outcome. The results are visualized in Figure 20. Second, we measured accessibility to gross floor area (GFA) around the same 14,218 Origin buildings, capturing variations in built density around each potential retail site. Here, we used the same set of Destination points as the previous analysis but weighted them with GFA instead of residents. The results are visualized in Figure 21.
URBAN NE T WORK ANALYSIS
Third, we also measured access to subway stops, using the same Origin buildings but changing the Destinations to subway stops. Each stop was weighted with a simple “Count” variable, which is equivalent of assigning it a weight of “1”. Again, surrounding stops that fell within a 600m buffer of Cambridge and Somerville were included as Destinations to avoid an artificial edge effect. The results are visualized in Figure 22. The three maps illustrate how spatial accessibility to residents, GFA and underground rail transit varies across the two towns. Values attached to individual buildings could constitute independent variables for a retail location choice regression model. 2. 3 . 2. 2.
Betweenness Analysis
While the Gravity Index captures the ease with which stores can be accessed from neighboring homes or workplaces, it does not tell us much about the ease of accessing stores en route, while traveling between other locations. Externalities and spillover effects can make a convenience store more desirable if it is not located at a place that is closest to people’s homes or jobs, but rather at a place, where people tend to pass by while traveling between other Destinations. We estimated the potential of passersby at different store locations using the Betweenness tool in the UNA toolbox. Ideally, this would be done by performing the Betweenness analysis from all buildings to all buildings within the boundaries of the two cities. Since the buildings in Cambridge and Somerville vary considerably in size, the Betweenness measures should be weighted by building volume — assuming that the number of trips originating from each building is proportional to the size of the building, with larger buildings emitting more trips than smaller buildings. The analysis should also be performed without a radius constraint to reach all corners of the two cities, but applying a different decay rate that would account for transportation costs. Such a Betweenness estimate thus assumes that the number of trips originating from each building is proportional to the size of the building, with larger buildings emitting more trips than smaller buildings. However, such ideal Betweenness analyses often involve too many calculations to be finished in a practical computing time. We therefore illustrate a somewhat simplified specification here, which is more practical in many contexts. Instead of performing the Betweenness analysis from all buildings to all buildings,
49
50
cambridge and somerville, ma
Figure 20: Gravity access to residential population within a 600m radius. (Gravity index of individual buildings weighted by residential count, r=600m)
Medford Arlington
Key Somerville
5455 5455 Gravity Access to 2156 2156Residential Population
00
Cambridge
Belmont
Gravity Access to residential population
Watertown
Boston
Figure 21: Gravity access to building volume within a 600m radius. (Gravity index of individual buildings weighted by GFA, r=600m)
Medford Arlington
Key 5455 304 404 Gravity Access to 2156 155 035 Residential Population
Somerville
00 Gravity Access to building volume
Cambridge
Belmont
Watertown
Boston
URBAN NE T WORK ANALYSIS
51 Figure 22: Gravity access to subway stops within a 600m radius. (Gravity index of individual buildings weighted by availability of subway stops, r=600m)
Medford Arlington
Davis
Key Key
Alewife Assembly Porter
Somerville
1.00 1.00 5455
GravityAccess Accesstoto Gravity 0.50 2156 0.50 Residential Subway StopsPopulation
Cambridge Belmont
0.00 0 0.00
Harvard
Central
Watertown
Subway stops Subway stops
Lechmere
Gravity Access to Subway Stops
Kendall/MIT
Boston
Figure 23: Betweenness from all intersections and ends to all intersections and ends, weighted by building volume (GFA).
Medford Arlington
Key 5455 212 004 802 Gravity Access to 2156 53 148 927 Residential Population
0 2180 Belmont
Betweenness
Watertown
Boston
52
cambridge and somerville, ma
we used all “intersection points” and “end points” of the street network as Origins as well as Destinations and spatially joined the nearest building points to them in Arc GIS, summing the GFA values. Shifting from individual buildings to individual street intersections thus decreased the spatial granularity of the analysis, but both intersections and street end points still carried weight values that represented the sum of building GFAs closest to them. The simplified spatial units of analysis still carried the same total gross building floor area weights. The Betweenness results are visualized in Figure 23. Additionally, in order to account for trips going in and out of Cambridge and Somerville, we also attached weights to the points along those, where the streets and bridges leading out of both towns interected with the town boundaries. These weights describe how much floor area in the outlying urban districts is closest to each departing street in a one-kilometer network radius. To compute these weights, another UNA tool — Closest Facility — was deployed. The Closest Facility tool was used to calculate the sum of weights that were closest to each departing street “midpoint”, avoiding double-counting that would have otherwise occurred with a Reach or Gravity accessibility index. The departing street “midpoints” were included in the final Betweenness analysis, with their weights corresponding to all the unique GFA that lay behind them in a one-kilometer buffer. These “gateway streets” were thus assumed to emit a much greater number of trips than a typical building. The resulting Betweenness values indicated the estimated number of passersby in front of each potential shop location (Figure 23). In order to control for immediate site characteristics, we also measured the median household income, vacancy levels, the proportion of renters and the proportion of elderly in the corresponding census tract of each building; the width of the road and the width of the sidewalk fronting each building; and parcel type — a variable that captures the number of streets that a parcel can directly access. While a “middle parcel” has access to a single street, a “corner parcel” or a “through parcel” faces two streets, an “end parcel” three streets, and so on. The ground floor area of each building was used as a proxy for building typologies. Small building footprints typically correspond to wooden structures in Cambridge and Somerville, while large footprints are more typically commercial structures or multistory apartment buildings that may have commercial spaces on the ground floor. All these indicators were combined as independent variables in a regression
URBAN NE T WORK ANALYSIS
model, while the binary dependent variable indicated whether (1) or not (0) the 14,218 buildings that constitute the units of analysis, contained any retail or food businesses. The Accessibility Indices, Betweenness values as well as site characteristics were effectively used to predict the probabilities of observing businesses at different locations. The results are presented in Table 2. 2. 3 . 3 . RESULTS
Regression coefficients in Table 2 illustrate whether and how the aforementioned location and site characteristics, as well as other variables used in the model, explain buildings’ probabilities to contain retail establishments. Gravity accessibility variables told an interesting story. Accessibility to subway stations was significant and positive (p < 0.0001). Holding other variables constant, a typical building that does have a subway station within a 600-meter (1,968 feet) network radius, is 2 percent more likely to accommodate a retail or food service business than a building that has no access to subway within 600 meters, controlling for covariates. This effect increases to 5 percent if there is a subway station within 100 meters (328 feet) from the building. As expected, retailers gravitate towards subway stations — a significant proportion of their patrons are likely to arrive by transit, particularly in the larger clusters, such as Harvard Square. Access to GFA was also positive (p < 0.001) and significant, suggesting that the likelihood of retail and food establishments increases in buildings that have more built volume in a ten-minute walking radius around them. Holding other variables constant, a building in a location with dense built volume around it (95th percentile) was 3.8 percent more likely to contain retailers than a building with a low-density built volume (5th percentile) around it. Surprisingly, the coefficient for residential access was negative (p < 0.0001), suggesting that retail and eating establishments tend to locate farther from areas of residential concentration. Keeping all other variables unchanged, a typical location with high residential access was 2.1 percent less likely to contain retailers than a location with low residential access. Even when taken alone, Gravity access to residents only exhibited a mildly significant positive relationship to store locations in Cambridge and Somerville. This counterintuitive residential effect could be explained by zoning, which keeps shops and restaurants out of more residential areas, as well as the fact that residential and employment distribu-
53
54
cambridge and somerville, ma
Table 2: Results of regression analysis.
Variable
Spatial lag estimates for location choice variables of retail and food establishments in Cambridge/Somerville, MA (n=14,218). Binary dependent variable: presence (1) or lack (0) of retail/food establishments in each building. Significance level ~ p<0.25 * p<0.1 ** p<0.05 *** p<0.01
R2
ρ (lag)
Coefficients
t and z statistics
0,28***
17,45
-1,458E-01***
-12,85
Subway stops (Gravity, r=600m)
6,210E-02***
5,66
Built Volume (Gravity, r=600m)
3,004E-09***
Constant
Residents (Gravity, r=600m)
-6,686E-06**
3,41 -2,37
Employees (non-retail or food, Gravity, r=600m)
9,813E-07
Betweenness (Weights=building volume, r=n)
3,254E-14***
10,38
Parcel Type (# of streets directly accessed 1-5)
8,477E-02***
32,45
Building footpring area (1000s of sq feet)
1,579E-07***
3,73
Road width
5,276E-04*
1,92
Sidewalk width
2,418E-03**
% Vacant % over age 60 Likelihood ratio test for spatial dependence
0,67
2,36
-1,324E-01~
-1,31
9,367E-02*
1,93
0,147 313,720***
tion are not entirely independent — as more space is taken up by jobs, less is left for residents and vice versa. Correlations between residential access and other access variables show that values for residential access are strongly correlated with access to building volume and tend to cancel each other out. The negative effect of residential accessibility is also likely to be specific to the fine-grain building resolution of our analysis. At a coarser resolution, such as at census-tract or zip code level, retailers indeed appear positively attracted to areas with more residents. Overall, Gravity accessibility results to subways and built volume confirmed that locations with better access to employment and transit Destinations have higher probabilities to host stores. Betweenness values were highly significant and positive (t= 10.38). This confirmed that the probability of observing retailers is significantly higher at locations that lie on more trafficked routes between other buildings. A high Betweenness (95th percentile) location was 5.88 percent more likely to host a store than a low Betweenness location (5th percentile), holding all other variables constant. The analysis thus showed that retailers were drawn to not only places where they can be accessed more readily from immediate buildings, but also places that accommodated more passersby. 2. 3 . 4 . APPLICATIONS
The Cambridge and Somerville retail location example shows
URBAN NE T WORK ANALYSIS
how UNA Accessibility Indices (Gravity Index in particular) and Betweenness tools can be used in spatial analytic models to explain or predict business locations. An interesting side benefit of such regressions, is that they also output residuals — positive or negative deviations between actual data points and the best guess trend line. Buildings with large negative residuals illustrate locations, where no store is currently present, but where the model expects stores to locate. These locations can help analysts find potential locations for new businesses. Strong positive residuals, on the other hand, can help detect locations where stores exist and operate despite poor location advantages. Besides the accessibility metrics and a particular application of Betweenness analysis illustrated here, the UNA Accessibility Indices and Betweenness tools can also be used with different Origins, Destinations and “weights” to explore how other spatial accessibility characteristics may affect or explain business or land use location patterns. For instance, a similar methodology can be applied to study design firm locations patterns, financial establishment clustering, affordable housing location patterns and so on, with appropriate specifications to suit each context. The next section turns to the nuts and bolts of the UNA toolbox itself, providing a detailed explanation of all available analysis functions and user controls.
55
56
URBAN NE T WORK ANALYSIS
3. DOWNÂLOADING AND INSTALLING
57
58
cambridge and somerville, ma
URBAN NE T WORK ANALYSIS
3 .1 .
DOWNLOADING
Before installing the UNA toolbar, please make sure you have Rhino version 5 or higher and the latest updates from McNeel installed. Updates can be installed by opening Rhino and going to help > check for updates. In order for the UNA tools to work, you need to use Rhino on the Windows operating system. You also need to have Dot Net Framework 4.5.1 or later on your windows operating system (free download from Microsoft). If you use Windows 10 or later, this should be automatically included and no extra updating should be required. Download the UNA Toolbox folder from one of the following following links: City Form Lab website: http://cityform.gsd.harvard.edu/ projects/una-rhino-toolbox Bitbucket: https://bitbucket.org/cityformlab/urban-network-analysis-toolbox/downloads/ Food for Rhino: http://www.food4rhino.com/app/urban-network-analysis-toolbox
59 Figure 24: Harvard GSD studio trays.
60
cambridge and somerville, ma
URBAN NE T WORK ANALYSIS
3 . 2.
INSTALLING
61 Figure 25: Grid-structure installation at the GSD.
In the download folder, you should see the “UNAToolbox.rhi” file. Drag and drop the .rhi installer file into an open Rhino window. This is the recommended approach. You can also double-click on the .rhi file and navigate through installation prompts, but note that some McNeel updates block the .rhi installer from being recognized on Windows. This is a known Rhino bug that McNeel is aware of. You can still install the UNA toolbox by dragging and dropping the *.rhi installer into an open Rhino window or re-associating the *.rhi file with C:\Program Files\Rhinoceros 5 (64-bit)\System\x64\rhiexec.exe. Once installed, restart Rhino and go to Tools > Toolbars tab and make sure to check the box next for the UNA Toolbar. Then Click “OK” to close the Toolbar window. You should now see the UNA toolbar appear in Rhino. Figure 26: Adding UNA toolbar in Rhino Options
Figure 27: UNA toolbar
The toolbar contains the tools that enable Urban Network Analysis in Rhino. You can also activate each of the tools from the command line. UNA tools typically start with “una…” followed by the tool name for each function. For example, typing "unaGraphics" on the command line opens the UNA Graphic Options menu.
62
cambridge and somerville, ma
URBAN NE T WORK ANALYSIS
4. GRAPHIC USER INTERFACE
63
64
cambridge and somerville, ma
URBAN NE T WORK ANALYSIS
65 Figure 28: UNA toolbar
The GUI is divided into five sections, which are visually divided by small vertical lines in the toolbar: attribute editing, data import/ export, network setup, analysis, and graphic settings functionality respectively.
4 .1 .
MANIPULATING ATTRIBUTES
Figure 29: Street signs in the Mong Kok district, Kowloon, Hong Kong.
The first set of the tools (first six from the left) deal with assigning objects attributes, which are optional to any analysis. The UNA Toolbox for Rhino utilizes “attributes” to attach information to Origin or Destination objects, such as text labels, numeric weights or tags. These attributes are analogous to attribute tables in ArcGIS shapefiles or BIM geometry files. An object can have any number of attributes. However, unlike some other software platforms, Rhino attributes are not stored as a table, but rather as a dictionary, where the key indicates the attribute name and values indicate attributes values. Attributes can be numeric, text based or booleans. Attributes allow you to distinguish spatial objects from one another by giving them unique properties that correspond to their real-world or designed characteristics. 4 .1 .1 .
ADD TEXT ATTRIBUTE
Assigning a Text Attribute to geometric objects in Rhino can be used to designate a “Business_name” or “Cleanliness_Rating” in the range of “A”, “B”, “C” and so on. Text attributes you enter can also be used as IDs, addresses or any other descriptive fields. Only numeric attributes can be used as weights in UNA analyses (e.g. in Accessibility Indices, Betweenness, Closest facility or Patronage tools). The Add Text Attribute tool allows you to select one or more objects at a time and then prompts for an attribute Name and attribute Text. Your chosen Name is a dictionary key, analogous to a column name in a table. Names cannot contain spaces. Text is
Figure 30: Add Text Attribute tool
66
manipulating attributes
the value that corresponds to the attribute (e.g. dictionary value or row value in a table). 4 .1 . 2. ADD Figure 31: Add Numeric Attribute tool
This tool allows you to add a numeric attribute to geometric objects in Rhino. Numeric attributes are particularly useful for spatial network analysis as they allow you to “weigh” the analysis according to each object’s numeric value. A numeric attribute could, for instance, indicate the “Number of Floors” in a building, the “Area” of a space, the “Number of Employees” in a business etc. The tool allows you to select one or more objects at a time for inputting these attributes and then prompts for an attribute Name and an attribute Weight. Your chosen Name is dictionary key, analogous to a column name in a table. Names cannot contain spaces. The Weight is a numeric value that corresponds to this Attribute (e.g. dictionary vlaue or row value in a table). The numeric input may be an integer or a decimal point number, negative or positive. 4 .1 . 3 . ADD
Figure 32: Add Tag Attribute tool
NUMERIC ATTRIBUTE
TAG ATTRIBUTE
This tool allows you to add a tag to chosen objects. Tags are booleans descriptions, with only True or Null values. For instance, attaching a tag named “building” to a point, will only show a value “Building = True” in the object attributes, with no text or numeric dictionary value. After selecting the objects, the tool prompts you for a tag Name on the command line. Tag names must contain text, they cannot consist of only numbers. 4 .1 . 4 . REMOVE
Figure 33: Remove Attribute tool
ATTRIBUTE FROM OBJECTS
This tool allows you to get rid of previously added attributes from chosen objects. If you choose all objects in the scene and apply this tool, then all objects will lose the text, numeric or tag names you specify. After selecting the objects, the tool prompts you for the attribute name to delete by asking Pick or type attribute name (AttributeName = …). Type in the tag name that you want to remove or click on the AttributeName link to choose available attributes. Note that you cannot remove GUID, which is a unique ID for each object, automatically assigned by Rhino.
URBAN NE T WORK ANALYSIS
4 .1 . 5 . SAVE
67
RESULT AS WEIGHT
This tool allows you to save Accessibility or Betweenness results to object attributes, which are by default just shown on the screen. The tool also allows you to assign the resulting attributes a custom name. In order to use the tool, you need to select point objects that contain some calculated results. Note that if you ran the analysis, and turned off the original point object layer, the results can still appear from cache memory on the screen. You cannot select the colored result points that Rhino displays on the screen and need to instead turn on the original points and select them. Once you have selected points with results, the UNA command line prompts you to choose a Weight name to designate the result you want to save by clicking on Result = … If you choose to save the results of your Gravity accessibility analysis, for instance, then click on Result = … and then choose Gravity. UseDefault = Off option on the command line allows you to turn on or off default values for points that do not have any results. For example, some points in your selection might not have Gravity results attached to them, since they could have been outside of the specified Search Radius. If you keep UseDefault = Off, then these points will have blanks as Gravity results. If you turn UseDefault = On, then you can assign default values to points with no results, by inputting a value on the DefaultValue = … option in the command line. For instance, if you want to assign all points that lack Gravity results a zero value, you can set DefaultValue = 0. Override = On option enables you to dictate whether existing values with the same name — if they exist — are over written or not. Finally, to assign a name for the weights you save, just type a name to the end of the options on the command line. It is good practice to use names that remind you of the inputs that were used in the analysis. For Gravity accessibility results that were run with a 3,000-meter radius and a beta value of 0.002, you could use a name like “Gravity_3000m_b002” or similar.
Figure 34: Save Result as Weight tool
68
manipulating attributes
4 .1 .6 . SHOW Figure 35: Show Attribute Tree tool
Figure 36: Attribute tree
ATTRIBUTE TREE
This tool allows you to see what attributes objects contain. The attributes are structured as a tree, showing the name of the attribute, value type (e.g. string, double, integer etc.) and value. Note that every geometric object in Rhino automatically carries a unique GUID. This is the first attribute you see in any object’s Attribute Tree. Figure 36 shows an Attribute Tree of a point, containing a GUID, a Text Attribute called “Name” with a value “Willow_Ave”, a Numeric Attribute called “Area” with a value of “13106.5972” and a boolean Tag called "Available". The “Load Selected” button allows you to pull up the attributes of only selected objects. The “Load” button lists the attributes of all objects in your Rhino scene. Note that if you have a lot of objects, this list can be very long and hard to follow, loading selected attributes is typically more practical.
70
manipulating attributes
URBAN NE T WORK ANALYSIS
4 . 2.
IMPORT EXPORT FUNCTIONS
71 Figure 37: Keppel Terminal in Singapore.
This section of the toolbar contains tools that allow you to import or export data from UNA network objects to .csv or .tsv files to or use in other software, such Excel, ArcGIS, Python etc. 4 . 2.1 . IMPORT
POINTS
Import Points tool can be used to import Origin or Destination point data with attributes from Excel, other text or table files or GIS shapefiles into Rhino. The tables you import should include X, Y and optionally Z coordinates, which will be used to draw the points in Rhino. Other attribute columns from the original table are also brought along as points attributes in Rhino. Note that there are some restrictions as to how your table columns can be named so that Rhino can recognize them as attribute names. Spaces in column names are automatically ignored (e.g. “My Name” is converted to “MyName”). Underscores “_” that form the first character in a column name are automatically removed. Column names can contain numbers but the entire name may not be a number. A few of names are reserved for internal UNA processes — for instance, “none” and “count” may not be used as column names. The command line prompts for several options. The Format = tsv option indicates that the imported file must be in .tsv (tab separated value) format. The most common way to save tables with point coordinates in tsv format is to first use Excel to save a table as a “Text (tab delimited) *.txt” extension. Then close the file in Excel, navigate to the saved location and rename the file extension manually from “.txt” to “.tsv”. Ignore the Windows warning. For some windows users, common file extensions, such as “.txt” are hidden by default. To rename the extension, you first need to turn the extensions on, so they are visible. On Windows Explorer, go to View tab and check the box next to “file extensions”. Click the View tab in File Explorer and then click the Options button (or click the drop down menu and click Change folder and search options).
Figure 38: Import points tool
72
import export functions
Figure 39: Click the Options button in the View tab.
Click the View tab at the top of Folder Options to see file extensions, uncheck Hide extensions for known file types. Click OK. Figure 40: Change viewing settings in Folder options.
This .tsv saving procedure is best done in the Windows version of Excel. The Mac version of Excel can produce a different encoding for “Text (tab delimited) *.txt” files, depending on the version of Excel used. This can result in “Text (tab delimited) *.txt” files saved on Mac not being able to be read by Rhino, even after changing the file extension to “.tsv”. If this problem occurs, then one workaround to still properly convert Mac Excel tables to “Text (tab delimited) *.txt” and subsequently to “.tsv” files, is to use a free text editor, such as Notepad++ to change the Mac encoding to Windows encoding. Download and install Notepad++ and open the “Text (tab delimited) *.txt” file that you saved from Mac Excel. Go to Edit>EOL Conversion > and set the option to “Windows (CR LF)”. Close the file in Notepad++ and rename its extension from “.txt” to “.tsv”. Notepad++ also allows you to save the file directly with a “.tsv” extension if you overwrite it in the “Save As” dialog box. The Import Points tool offers several options on the command line: File option enables you to navigate to the “.tsv” file you want to import. Attributes = On option prompts the UNA toolbox to either import points with or without tags. If used as Attributes = Off,
URBAN NE T WORK ANALYSIS
73
then points can be generated in Rhino without any tags included from the text file. Weight = … option indicates whether or not numeric attributes are included or not in the import. Text = … option indicates whether or not text attributes are included or not in the import. InvalidCoordinate = … option determines how the import procedure should handle invalid X, Y or Z coordinates. The default option sets such invalid coordinates to zero. Once the inputs are set, press enter. The next step on the command line asks you to indicate which field in the input table contains X, Y and Z coordinates. Click on each coordinate to assign a corresponding column header from the imported table. Note that the Z coordinate is optional, but both X and Y coordinates are required. Pay attention to the units of measurement. If the Rhino drawing uses “meters” as units, then the imported X and Y coordinates should also use “meters”, so that the created points fall in the right location. If the Rhino drawing uses “feet”, imported X and Y should also be in “feet”. A common method for creating X and Y coordinates for points imported from shapefiles is to use the Calculate Geometry function for points in ArcGIS. More information on that can be found on ESRI support sites: http://desktop.arcgis.com/en/ arcmap/10.3/tools/data-management-toolbox/add-geometry-attributes.htm 4 . 2. 2. IMPORT
TABLE
This tool allows you to bring in a table of attributes that will be joined to existing point objects based on their GUID (Rhino assigned object IDs). This tool allows you to bring new attributes data that was created outside of Rhino using other tools (e.g. Excel), back to objects, thereby taking advantage of powerful formulas, query functions and calculation tools that are not available within Rhino. All objects in Rhino contain an automatically assigned GUID, which is a 36-character unique ID, containing of both number and letter characters, for example: “6a09b867-3cf1-402e-a7fd31008e4ddf5f”. When you use the Export tool in the UNA toolbar (discussed below), a GUID is always included as part of the object data that is exported. By keeping this GUID field and adding, subtracting or editing other data columns in the table, you can
Figure 41: Import Table tool
74
import export functions
bring newly computed attributes back into Rhino. The new data will be joined to the same objects, based on GUIDs you exported in the first place. Alternatively, you can see Rhino object GUIDs by a) exploring object properties > details, or b) using the UNA Show Attribute Tree function. Format = … option on the command line enables you to choose whether the imported table is in a .tsv or .csv format, both are allowed. Note that Excel can save tables into .csv format, without having to do any further file extension renaming, which is required when you use the .tsv format. File option allows you to navigate to the desired text file to import. Update = All option determines whether all or only selected object attributes are updated as part of the import. Override = Off function determines whether or not object attributes that already exist in Rhino with the same attribute name are replaced or not. For instance, if you already have an attribute called Reach on your point layer while the imported table also contains an attribute field called Reach, and you set Override = On, then the Reach attributes in Rhino will be over written with new Reach values from the imported table. 4 . 2. 3 . EXPORT Figure 42: Export tool
The Export tool allows you to export existing UNA object attributes to a table. These could include UNA analysis results (e.g. Accessibility, Betweenness etc.), as well as text, numeric or tag attributes you have manually created in Rhino or imported from other tables. The exported information is stored in the form of a table, which can either be saved to a chosen file location or copied to clipboard memory. In the latter case you can subsequently paste the exported table into other software, such as Excel or a text editor. The tool offers a number of options on the command line. Format = tsv option determines in which format the output table is stored. While the default is Tab Separated Values, which is easily compatible with Excel, clicking on the option opens up the following possible table export formats: Tab Separated Values (tsv), Comma Separated Values (csv), Separator Separated Values (ssv), and Geojson. Export type = memory option allows you to choose whether the table is exported to a new file or into clipboard memory. If you choose file, a location is prompted.
URBAN NE T WORK ANALYSIS
Points = include option enables or disables points attributes from the table. It is on by default. Most often the export function is used to simply export results of UNA Origin, Destination or Observer point objects. Curves = Exclude option enables or disables curve attributes from the table. It is off by default. Even though attributes can be added to curve objects, the UNA toolbox does not compute any results for curve objects. The Betweenness tool can display results at the curve level, but this is an approximation — average of endpoint Betweenness results — that cannot be exported. Results = Exclude option determines whether UNA analysis results, such as Accessibility or Betweenness results, which have not been saved as weights yet, are included in the table. The option is off by default, but if turned on, it enables latest analysis results to be exported directly, without having to first save the results as weights. Attributes = Include option determines whether saved objects attributes, which are not current analysis results, are included in the export. It is turned on by default. Type = Exclude option determines if the object Type is included in the table. This can be used to distinguish points from curves, for instance. The option is excluded by default. Coordinates = Exclude determines whether or not the X, Y and Z coordinates of the exported objects are exported. If turned on, the exported points and their attribute data can be reconstructed from the table in other software, such as ArcGIS, CartoDB, OpenStreetmap, MapBox, Processing, etc. EPSG = 3857 option sets the coordinate system used in creating the X, Y and Z coordinates. By default, the “3857” option uses the “WGS 84 / Web Mercator — Spherical Mercator” projection, that is commonly used by Google Maps, OpenStreetMap, Bing, and ArcGIS.
75
76
import export functions
URBAN NE T WORK ANALYSIS
4.3.
NETWORK CREATION AND EDITING
77 Figure 43: Pedestrian paths behind Harvard's Gund Hall.
Next, we turn to tools that allow you to construct spatial networks, add or remove Origins and Destinations on the networks and delete networks. 4 . 3 .1 . ADD/REMOVE
CURVES FROM NETWORK
Any network analysis in the Rhino UNA toolbox requires at least Figure 44: Add/Remove two geometry inputs — first, an input analysis network along which Curves tool all trips are computed, and second, Origin and Destination points, where movement starts or ends. The Add Curves to Network tool prompts you to select curves that you want to build your network out of. Select all the curves you want to involve in the network and press enter. This will automatically turn the selected curves into a network and build an adjacency matrix that is used for analysis. Note that through a left click, the tool also enables you to add curves to an existing network. If some of the curves you might be No nodes at interadding to a pre-existing network are already part of the network, section: no conno nodes at intersection: no continuity tinuity betweenbetween 2 cur their GUIDs will be recognized and they will not benodouble nodes atrepintersection: no continuity between 2 cu 2 curves no nodes at intersection: no continuity between 2 cu resented. Right clicking the tool, on the other hand, allows you to remove curves from an existing network. In order for network analyses to work, network curves need to provide continuity between the Origin and Destination points you analyze (see Figure 45). If your network curves do not share 1 node at intersec1 node at intersection:tion: no continuity between 2 curv no continuity a common end node, then Origin and Destination points found 1 node at intersection: no continuity between 2 cur between 2 curves 1 node at intersection: no continuity between 2 curv on different network segments might not be topologically connected to each other. If one curve ends on top of another, but the latter does not have a node at the intersection (e.g. T intersection with no shared endpoints), then there is no topological continuity between the two curves. Curves that intersect without sharing a 3 continuity nodes at inter3 nodes at intersection: between all 3 curv section: continu3 nodes at intersection: continuity between all 3 cur common node can be used to model three-dimensional overpasses 3 nodes at intersection: all 3 cur itycontinuity between between all 3 or underpasses. curves The tool can accept any kinds of curves to participate in Figure 45: networks — lines, polylines, curves, arcs, etc. Networks of these Nodes and continucurves can either be planar (2D) or three-dimensional, as long ity between curves
78
net work creation and editing
2
Figure 46: A “Degree 2” node has exactly two curves intersecting at the node.
as adjacent curves share a common end node with each other. While 2D networks may be adequate for representing street and building networks in urban settings, 3D networks offer additional opportunities for analyzing circulation systems and layouts within buildings or in multi-layered urban infrastructure systems. For visual clarity, the default settings in the UNA Graphic Options will visualize dead-ends or “naked edges” of the network with little black crosses (Figure 46). You can turn these black crosses off in the Graphic Options by turning off Nodes. The black crosses can be useful to visualize where your network might contain topological errors. In Figure 46, for instance, the first and second intersections from the top have topology issues. On the first intersection, a black cross is drawn, indicating that one or more curves around that node has a dead end and does not connect to any other curve. On the second node, a red warning with a numeric value “2” is displayed. This warning, which can also be toggled on or off in the Graphics Options using the NodeD2 setting, detects “degree 2” nodes — that is nodes that have exactly two curves intersecting at the node. In this case it signals two polylines meeeting at right angles, instead of four segments sharing an endpoint. 4 . 3 . 2. ADD/REMOVE
Figure 47: Add/Remove Origins tool
ORIGINS
The Add Origins tool adds trip Origin locations to the network. As with network lines, left-clicking adds Origin points, right-clicking removes Origin points. All UNA analysis functions require both Origin and Destination points, which can designate any spatial locations, such as address points, buildings, entrances, rooms in a building or even locations in utility or infrastructure networks, such as airports, stations etc. Analysis results are typically computed for either Origin or Destination points (and sometimes observer points), depending on the UNA tool used. For instance, Accessibility results, such as Reach or Gravity metrics, are always returned for Origin points, while Closest Facility results are instead returned to Destination points. The tool first asks for a selection of points. After selecting points graphically (or right-clicking on a layer > Select Objects), the following options appear on the command line: Press Enter to add origins to network (Search=2D SavedEdges=On):
URBAN NE T WORK ANALYSIS
Search = 2D option determines whether the added points are snapped to the network in two-dimensional (default) or three-dimensional space. For planar networks, where all network links are on the same level (as is the case with many urban street network datasets), it is advisable to use 2D search, which functions faster than 3D search. When three-dimensional networks, such as building circulation networks, or multi-level urban networks are used, the Search option should also be set to 3D. SavedEdges = On option allows you to check if a point has a saved edge that it is designated to. It is possible to tie a point to a particular edge of the network, which may not necessarily be the closest edge, using the unaBindEdge function from the command line. If the edge reference that a point carries is not found in the drawing file, regular closest edge joining is used. You can typically ignore this input. Once you have added Origin points, each point is associated with its closest network element with a blue connection line. The location at which this blue connection line snaps to the original network, designates the assumed network location of the point. You can toggle the blue connection lines on or off in the Graphic Options by changing the DotConnections = On option. 4 . 3 . 3 . ADD/REMOVE
79
Figure 48: Origin points are tied to the network with blue dot connections
DESTINATIONS
The Add Destinations tool adds trip Destinations to the network. Left-clicking the tool adds Destinations points to network, right-clicking removes Destinations points from network. The tool first asks for a selection of points. After selecting points graphically (or right-clicking on a layer > Select Objects), the following options appear on the command line: Search = 2D option determines whether the added points are snapped to the network in two-dimensional (default) or three-dimensional space. For planar networks, where all network links are on the same level (as is the case with many urban street network datasets), it is advisable to use 2D search, which functions faster than 3D search. When three-dimensional networks, such as building circulation networks, or multi-level urban networks are used, the Search option should also be set to 3D. SavedEdges = On option allows you to check if a point has a saved edge that it is designated to. It is possible to tie a point to a particular edge of the network, which may not necessarily be the closest edge, using the unaBindEdge function from the command
Figure 49: Add/Remove Destinations tool
80
Figure 50: Destination points are tied to the network with red dot connections
net work creation and editing
line. If the edge reference that a point carries is not found in the drawing file, regular closest edge joining is used. You can typically ignore this input. Once you have added Destination points, each point is associated with its closest network element with a red connection line. The location at which this red connection line snaps to the original network, designates the assumed network location of the point. You can toggle the red and blue connection lines on or off in the Graphic Options by changing the DotConnections = On option. 4 . 3 . 4 . ADD/REMOVE
Figure 51: Add/Remove Observers tool
Figure 52: Observer points are tied to the network with grey dot connections
OBSERVERS
Observer points are only used in Betweenness analysis. They enable you to return Betweenness results for points that are neither Origins nor Destinations themselves, but rather locations that are potentially passed by trips between other Origins and Destinations. If trips originate from bus stops and go to shops, for instance, then building points in between bus stops and shops can be used as Observers to illustrate how many trips pass by each building. Betweenness analysis keeps track how many times each Observer point is passed in the analysis. The weights of Observer points are not used as part of any analysis. Observer points are tied to the network with gray dot connections, which indicate their network locations. Just like adding Origins or Destinations, left-clicking the tool adds Observers, right-clicking removes Observers from network. Note that Observer points can be particularly relevant if the Origin and Destination points are located on the same edge segment. Betweenness results for observer points are always exact, while the Betweenness values for edges are approximations, found by the taking the average of Betweenness values of the segment’s end nodes. Edge values can thus under-represent trips, where Origins and Destinations lie on the same network segment. Search = 2D option determines whether the added points are snapped to the network in two-dimensional (default) or three-dimensional space. For planar networks, where all network links are on the same level (as is the case with many urban street network datasets), it is advisable to use 2D search, which functions faster than 3D search. When three-dimensional networks, such as building circulation networks, or multi-level urban networks are used, the Search option should also be set to 3D.
URBAN NE T WORK ANALYSIS
81
SavedEdgesâ&#x20AC;&#x2030;=â&#x20AC;&#x2030;On option allows you to check if a point has a saved, designated edge that it is designated to. It is possible to tie a point to a particular edge of the network using the unaBindEdge function from the command line, which may not necessarily be the closest edge. If the edge reference that a point carries is not found in the drawing file, regular closest edge joining is used instead. You can typically ignore this input. 4 . 3 . 4 . DELETE
NETWORK
This function deletes the entire network representation you have previously added in the Rhino scene, including all networks, Origin, Destination and Observer points. Having deleted a network, you can start over, by designing a clean network with new Origin and Destination points. All analysis results will be deleted from memory. Only those analysis results that have been saved to object attributes using the Save Result as Weight tool will be maintained.
Figure 53: Delete Network tool
82
net work creation and editing
URBAN NE T WORK ANALYSIS
4.4.
ANALYSIS TOOLS
83 Figure 54: UNA workshop at ETH Zurich.
This set of tools in the UNA toolbar includes key functions that produce network analysis results — Accessibility Indices, Service Area, Redundant Paths, Betweenness, Closest Facility, Find Patronage, Distribute Weights and Clusters. 4 . 4 .1 . ACCESSIBILITY
INDICES
The Accessibility Indices tool launches the accessibility calculations between network Origin and Destinations points that have been added. The UNA toolbox computes three different Accessibility Indices — Reach, Gravity and Straightness — each of which offers a unique and complementary way of analyzing spatial relationships between Origin and Destination points in a network. In each case, results are returned to Origin points. If the analysis is weighted, weights are applied to Destination objects. If you would like to analyze accessibility from buildings to subway stops, for instance, then buildings should be taken as Origins and subway stops as Destinations. You can optionally weigh subway stops by a numeric attribute, attached to these stops, indicating how many metro lines or trains per hour each stop serves. Upon running the tool, a command line message will ask you to select the Origins for the analysis or to accept the pre-selection. The pre-selection automatically detects all Origin points you have previously added to the network. If you wish to use all of them for the analysis, then simply press Enter to accept the pre-selection. If you wish to only select a subset of points for analysis, then select those Origins you want to include. All accessibility results will be computed for the points you decide to select here. Selected Origins will be temporarily indicated with blue crosses. Next, the tool prompts you to select Destinations in the same way. Selected Destinations are marked with red crosses. Once you have selected Origin and Destination points, you will be prompted for a set of options on the command line. Search Radius <600> (Reach=On Gravity=On Straightness=On Weight=Count Beta=0.004 Alpha=1):
Figure 55: Accessibility Indices tool
Figure 56: Origin points are marked with blue crosses
Figure 57: Destination points are marked with red crosses
84
analysis tools
Search Radius input defines the network radius used for computing accessibility measures you choose. For each Origin point, only Destination points whose shortest network distance from the Origin is less than the specified Search Radius are considered in the analysis. Search Radius units follow the drawing units — if your drawing is in meters, Search Radius is also in meters. The active Search Radius is shown at the beginning of the command line prompt; you can change the Search Radius by typing a new number on the command line. Next, you can see a list of four Accessibility Indices that you can choose to include in the results: Reach = On option turns the Reach analysis on or off. Gravity = On option turns the Gravity analysis on or off. Straightness = On option turns the Straightness analysis on or off. Turn them on or off by clicking each one on the command line. Each of these Accessibility Indices is explained in detail below. Weight = Count option allows you to weigh the accessibility results with Destination attributes. When using Reach analysis to measure how many jobs one can walk to from each building in a ten-minute walkshed, for instance, you might weight the Destination buildings with a “Jobs” weight, indicating the number of jobs at each Destination building. Reach results would then illustrate the number of jobs that can be reached from every Origin building in a ten-minute walkshed. When the weights are kept as Weight = Count, then each Destination point is simply counted as “1” and no additional weights are applied. You can choose weights by clicking on the Weight = Count option, which will list all the numeric attribute weights that are available in the drawing. Click on the weight field you want to use and make sure that the weights you assign are actually associated with the Destination points in your network. See sections 4.1.2. Add Numeric Attribute, 4.2.1 Import Points, and 4.2.2. Import Table on how to assign weights to points. Beta = 0.004 and Alpha = 1 coefficients only affect the Gravity index. If you are not performing Gravity accessibility analysis and set Gravity = Off, you can disregard these inputs. The beta and alpha coefficients are explained as part of the Gravity index below. The next section discusses each of the four Accessibility Indices in detail.
URBAN NE T WORK ANALYSIS
4 . 4 .1 .1 .
85
Reach
The Reach index, also known as a “cumulative opportunities accessibility index” (Bhat 2000; Sevtsuk 2010; Jaber and Papaioannou 2017) captures how many surrounding Destinations (e.g buildings, businesses, jobs, bus stops etc.) can be reached from each Origin within a given Search Radius on the network. Reach returns a number with a prefix “r” on each Origin point. The Reach of an Origin i in a graph G describes the number of Destinations j in G that are reachable from i at a shortest path distance of at most r. It is defined as follows:
Reach[i]r =
W [j]
j∈G −{i},d[i,j]≤r
where d [i,j ] is the shortest path distance between Origin i and Destination j in G, and W [j ] is the weight of a Destination j. Figure 58 illustrates how the Reach index is calculated visually. A walkshed is traced from each Origin in every direction on the network until the limiting radius r is reached. The Reach index corresponds to the number of Destinations j that are found from the Origin within the Search Radius on the network. In order to simply compute the number of Destination points reached within a given Search Radius, set the Destination weights value as Weight = Count, so that no weighting will be applied and only the count of Destination buildings will be returned. To weight the measure by Destination weights, choose the weights to be applied by clicking on the corresponding Weights = Attribute value that is available in your dataset. For instance, you can give a gross floor area (GFA) attribute to Destination points and use the Reach measure to compute how much building GFA is reached within the Search Radius around each Origin on the network. To capture Reach to activities or land use Destinations, you can use the number of jobs, the number of residents, the number of business establishments and so on as Destination Weights. Figure 59 illustrates how many businesses can be reached from different buildings in Cambridge, MA in a 600m Search Radius. As expected, buildings closer to subway stations reach more businesses around them.
Equation 1
86
analysis tools
Figure 58: There are 24 other buildings that are within a 100m network distance from the building in blue.
r: 24 Origins: building in blue D estinations: red buildings Radius: 100m Weight: count
!
Figure 59: Buildings on Massachusetts Avenue and around the Central T station have higher Reach values to businesses within a 600m network distance.
!
!
!!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
! ! ! ! ! !
!
!
!
!
!
!
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
!!
!
!
!
!
!
!
! !
!
!
! !
!
922
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
! ! !
!
!
! ! !
! !
!
! !
!
! !
!
! !
! !
! ! !
!
! !
! !
!
!
!
!
! !
!
!
! !
! !
! !
! !
!
! !
! !
!
!
!
! !
!
!
! !
!
! ! ! !
!
! ! !
!
! !
!
!
!
! !
! !
!
! !
!
!
! !
!
! !
!
!
! ! !
!
! !
!
! !
! ! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
! !
!
! !
!
!
!
!
! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
! !
! !
!
! !
!
!
!
!
! ! !
!
! !
!
!
! !
!
!
!
! !
! !
! !
! !
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! ! !
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
! !
!!
!
! !
!
!
! !
!
!
!
! ! ! !
!
!
!
! !
!
! !
! !
!
! !
! !
! !
! !
!
!
! !
!
! !
!
! !
! !
! !
!
!
!
!
! !
!
! !
! !
!
!
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
! ! !
! !
! !
!
!
!
!
!
! !
! !
!
! !
!
! !
!! ! !
!
! !
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
! !
!
!
! !
!
!
!
! !
! !
!
!
! ! !
!
!
! !
! !
! ! !
! !
! ! ! !
!
!
! ! !
! ! ! ! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
! ! !
!
! !
!
!
!
! !
! !
!
!
! !
!
! !
!
! !
! !
!
!
!
!
!
!
! ! !
! !
!
!
! ! ! !
!
! !
!
!
!
! ! !
! ! ! ! !
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
!
!
!
!
!
! ! !
! !
!
!
!
!
!
!
!
! !
!
! ! ! !
!
! !
! !
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
! !
!
!
! !
!
! !
!
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! ! !
!
! !
!
!
!
! ! !
! !
!
!
!
!
! !
! !
! !
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
! !
!
!
! !
! !
!
! !
!
!
!
!
!
!
! ! ! ! !
!
! !
! ! !
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
! !! !
!
!
! ! ! ! ! ! !
! !
! !
! ! ! ! !
!
!
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
! !
! ! !
!
! !
!
!
!
!
! !
!
!
!
!
! !
! ! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! ! !
!
! !
!
! ! !
!
!
! !
!
! !
!
! !
!
! !
! !
!
!
! ! !
!
!
!
!
! !
! !
!
!
!
!
!
! !
!
!
! ! ! !
! !
!
!
!
!
!
!
!
! !
! !
! !
!
!
!
!
! !
! !
!
!
!
!
! ! !
!
!
! ! !
! !
!
!
! ! !
! !
! !
! !
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
! ! !
!
!
!
!
! !!
!
!
!
!
!
! !
! !
!
!
!
!
! !
! !
!
!
!
! !
!
!
!
!
!
! ! !
!
!
!
!
! ! !
! !
!
!
!
!
!
!
! !
!
! !
!
!
! !
!
! ! ! !
!
! !
!
! ! !
! ! !
!
!
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
!
! ! !
! ! ! ! !
! !
! !
!
! !
! !
! !
!
! !
! !
!
!
!
!
!
!
! !
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
!
! !
!
!
!
! ! !
!
!
!
!
!
! !
!
!
!
! ! !
!
!
! !
! !
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
! !
!
! !
! !
!
!
!
! !
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!!
!
!
! !
! !
!
!
!
! ! !
!
!
! !
! !
!
!
! !
! !
!
! !
!
!
!
!
! !
!
! !
!
!
!
!
!
! !
!
!
!
! !
! !
!
!
! ! !
!
!
!
!
! !
! !
! !
! !
!
!
! ! ! !
! !
! !
! ! !
!
!
!
!
!
!
!
! ! !
!
! !
!!
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
! !
!
!
!
!
! !
! !
! !
!
!
!
!
!
!
!!
!
!
!
! ! !
! !
!
!
!
! ! !!
!
!
!
!
!
!
! !
! !
! !
!
! !
!
!
!
! !
!
! !
!
! !
!
!
!
! !
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! !
! ! !
!
!
!
! !
! !
!
!
!
!
! !
!
! ! !
!
!
!
!
! !
!
! !
!
! !
!
!
!
!
!
! !
!
! ! !
!
!
! !
!
!
! !
! !
!
! !
! !
!
!
! ! ! !
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
! !
! !
!
! !
!
!
! !
!
!
!
!
!
!
!
! !
! !
!
!
!
!
Reach to businesses
!
!
! !
! !
! ! ! ! !
!
! !
!
! !
!
!
!
! !
! !
!
!
!
!
! ! !
!
!
!
!
!
! !
! !
!
!
!!
!
!
!
!
!
!
! !
! !
!
! !
!
!
! !
!
!
! !
!
!
!
!
! !
!
!
!
!
! !
!
!
!
!
! !
! !
!
! !
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
! !
! !
! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
! !
!
!
!
!
!
!
!
! ! !
! !
!
!
!
! !
!
!
!
!
!
! !
!
!
! ! ! !
! ! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
! !
! !
!
! !
!
!
! !
!
! !
!
!
! !
!
! !
!
!
! !
!
!
!
0
!
!
!
!
!
! ! !
! !
! !
!
!
!
! !
! ! !
!
!
!
!
! !
!
! !
!
! !
!
!
!
!
!
! !
!
! !
!
!
!
!
! !
! !
! !
! ! !
!
!
!
!
! ! !
! !
!
!
!
! ! !
!
!
!
!
!
!
! !
!
!
! !
! !
!
!
!
!
!
! !
!
! !
! !
!
!
! !
!
! ! !
! !
! ! !
!
!
!
!
!
!
!
!
!
! ! ! !
!
!
!
! !
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
! !
!
! !
!
!
!
!
!
461
! !
!
!
!
!
! !
!
!
!
! !
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
! !
!
!
!
!
!
!
!
! !
!
! !
!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! !
!
!
!
!
!
!
! !
!
!
!
! ! ! !
!
! !
!
!
!
! !
!
! !
! !
!
!
!
! !
! ! !
! !
!
!
!
!
! ! !
!
!
! !
! ! !! ! !! !
!
!
!
! !
! !
! !
! !
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!!
!
!
!
!
!
!
! ! ! ! !
! !
! !
!
!
! !
!
! !
!
!
! ! !
! !
!
!
!
!
!
!
!
!
! !
! !
!
! ! ! !
!
!
! !
!
! !
! !
! !
!
! !
! !!
!
!
!
!
!
!
! !
!
! !
!
! !
! ! !
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! ! ! !
!
!
! !
!
! ! !
! ! !
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
!
!
! !
! !
! ! !
!
!
!
!
!
!
!
!
!
! !
! !
! !
! !
! !
!
!
!
! ! !
! !
! !
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
!
! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
! ! !
! !
!
!
!
!
! ! ! !
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! ! !
!
!
!
! !
!
!
! !
!
! !
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
! ! ! ! ! ! !
!
!
!
!
!
!
! !
!
!
! !
!
! !
!
!
!
! ! ! !
!
!
! !
! !
!
!
!
!
!
!
!
!
! !
! !
!
!
!
! !
!
!
!
!
! ! ! !
!
!
!
!
!
!
! !
!
!
Radius: 600m Weight: count
!
!
!
!
!
!
! ! !
!
!!
!
! !
!
!! ! !
!
!
! !
!
!
!
!
! !
!
! !
!
!
!
!
! !
! !
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
!
! ! !
!
!
!
!
!
!
! ! !
!
!
! ! !
!
!
!
!
! !
!
! !
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
! ! !
! !
!
! !
!
!
!
! ! !
!
! !
! !
!
! !
! !
!
!
!
! !
!
!
! !
!
! !
!
!
!
!
! !
! !
! !
!
!
! ! !
!
!
!
!
!
!
!
!
!
D estinations: all businesses
!
!
!
!
! !
!
!
!
! ! !
!
!
! ! !
! !!
!
! ! !
! !! !! ! ! !
!
!
!
! !
!
!
!
! !
!
!
! !
!
! !
! !
!
!
!
!
! !
!
!
! ! !
!
! !
!
!
!
!
! ! !
! !
!
!
! !
!
!
! ! !
!
! !
!
!
! !
!
! !
!
! !
!
!
! !
!
!
!
!
!
!
! !
! !
! ! !
! ! !
! !
!
!
!
! !
!
!
!
!
!
! !
!
!
!
! !
!
! !
!
!
!
!
!
!
! ! ! !
!
!
!
! !
! !
!
!
!
!
! !
!
!
!
!
! !
!
!
!
!
! !
! !
!
! !
!
!
!
!
!
!
!
!
! !
! !
!
!
! !
! !
!
!
! !
!
!
!
! !
!
!
!
! !
! !
! !
! !
!
! !
! !
! !
!
! ! !
!
!
! !
!
!
! ! !
!
!
! ! !
!
!
!
!
!
!
! ! !
! ! ! ! !
!
!
! !
! ! ! ! !
!
! !
! !
!
! !
!
! !
!
!
!
! !
!
! !
!
!
! !
!
! ! !
!
!
! !
! !
!
!
!
!
!
! !
!
!
! !
!
!
!
! !
!
!
!
!
!
!
! !
! !
!
!
!
! !
! !
!
! !
!
!
! !
! !
!
! !
! !
!
! ! ! !
!
!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
! ! !
! !
!
! !
!
!
! ! !
!
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
! ! !
!
!
! ! !
!
! !
!
! !
!
! !
!
!
!
!
!
!
!
O rigins: all buildings
!
!
! ! !
! !
!
! !
!
! !
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !
!
!
!
! ! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
! ! !
!
!
!
!
! ! !
!
!
!
! !
!
! ! !
!
!! !
!
!
!
!
!
!
! !
!
! ! !
!
! !
! !! !
!
!
!
!
! !
!
!
! !
! ! !
!
! ! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
! !
!
!
!
! ! !
! ! !
!
!
!
!
! !
! !
! !
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
! !
! !
! !
! !
!
! !
!
! !
! ! !
! !
! !
! ! ! !
!
! !
!
!
!
!
! ! !
! ! !
! !
!
!
! ! ! ! !
!
!
!
!
! !
! ! !
!
! !
!
! !
! ! ! ! !
! !
!
!
! ! ! !
!! !
! ! !
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! ! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!! ! ! ! !
!
!
! !
! ! !
! !
!! ! !
! !
! !
!
!
! !
! ! ! !
! !
!
! ! !
! !
! !
!
!
!
!
!
! ! ! !
! !
!
! !
!
! !
!
!
! !
!! !
!! !!
! !
!
!
!
! !
!
!
!
! !
! ! !
!
!
!
!
!
!
!!
!!
! ! !
! ! !
! !
! ! !
!
! !
!
!
!
!
! !
!
!
!
! !
! !
!
! !
!
!!
! ! !
! !!
!
! !
!
!
!
!
! !
!
!
!!
!! !!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!!
! !
! !
!
!! !
! ! ! !!
!
!
!
! !
!
!
!
!!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
URBAN NE T WORK ANALYSIS 4 . 4 .1 . 2.
87
Gravity
Whereas the Reach index simply counts the number of Destinations around each Origin within a given Search Radius (optionally weighted by Destination attributes), the Gravity index additionally accounts for travel costs required to reach each of the Destinations. Gravity returns a number with a prefix “g” on each Origin point. First introduced by Hansen (1959), the Gravity index remains one of the most popular spatial accessibility measures in transportation research. Gravity index assumes that accessibility at Origin i is proportional to the attractiveness (weight) of Destinations j, and inversely proportional to the distance or travel cost between i and j. In the current version of the Rhino UNA toolbox, only network distance can be used as a measure of travel costs. The index is defined as follows:
Gravity[i]r =
j∈G −{i},d[i,j]≤r
Equation 2
W [j]α eβ·d[i,j]
1
beta = 0.002 beta = 0.004
0.9
beta = 0.02
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
1,900
2,000
1,700
1,800
1,500
1,600
1,300
1,400
1,100
1,200
900
1,000
700
800
600
400
500
200
300
0
0 100
Gravity
where Gravity[i]r is the Gravity index at Origin i within graph G at Search Radius r, W[j] is the weight of Destination j, d [i,j] is the network distance between i and j, α is the exponent that controls the Destination weight or attractiveness effect, and β is the exponent that controls the “distance decay” effect. The Gravity index thus captures both the attractiveness of Destinations (W[j ]α)
Distance (m)
Figure 60: Beta measures the “distance decay” effect
88
analysis tools
Figure 61: This analysis measures the Gravity accessibility from the two bus stops to all buildings within a 200m network distance. The Gravity accessibility values for the bus stops are smaller than their Reach values, because the former accounts for the “distance decay” effect. O rigins: bus stops D estinations: all buildings Radius: 200m Weight: count Beta: 0.004 Alpha: 1
g: 56.729
g: 45.501
and the spatial impedance of travel required to reach the Destinations (d[i,j ]) in a combined measure of accessibility. If no Weights are chosen, then the weight of each Destination is considered to be equivalent to their count or “1”. The α parameter controls how a change in Destination weights changes the outcome. The default value is set at “1”, so that the Destination weight has an exponent of “1”, effectively assuming that as Destination size increases, Gravity index also increases in a linear manner. It is recommended to use the default value when the actual relationship between Destination weights and accessibility outcomes in unknown. For particular Origin - Destination pairs, however, alpha and beta parameters can be empirically estimated. For instance, Sevtsuk and Kalvo (2017) discuss how alpha and beta parameters for retail accessibility were empirically estimated using survey data in Singapore’s public housing towns. Such estimation requires empirical data about people’s trips to selected types of Destinations (e.g. retailers), describing how far the Origins were from Destinations, what the “weights” (e.g. floor areas) of the Destinations were, and how frequently people undertook the trips. The values can differ by socioeconomic group. Wheaton and DiPasquale (Chapter 6, 1996) discuss how similar parameters were empirically estimated in a phone survey of retail visits in the Boston metropolitan area. The effect of distance is inversely proportional to the Gravity index and it decreases exponentially. As distance to Destinations grows, the Gravity index decreases. But the index decreases faster around shorter distances and slower around longer distances. The exact shape of the distance decay rate is controlled with the exponent β, specified in the command line when running the Accessi-
URBAN NE T WORK ANALYSIS !
! !
!
!!
!
! !
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!! ! ! ! ! ! ! ! ! ! ! ! ! !
! ! !
!
!
! !
!
!!
!
!
! !
!
!
!
!
! ! ! ! ! !
! ! !
!
!
! !
! !
! ! !
! !
!
!
!
!
!
!
!
!! !
!
!!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
!
!
! !
! ! !
!
!
!
!
!
!
! !
!
!
! ! ! !
!
!
! !
!
! !
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
! !
! ! !
!
!
!
!
! !
!
!
!
!
!
! !
!
!
!
! !
! !
! !
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
! !
!
Gravity accessibility to transit stops !
!
!
!
!
!
!
!
!
!
!
!
! ! ! ! !
! !
! !!
!
! !
! !
! !
!
!
!
!
!
!
! !
!
! !
!
!
!
!
! !
!
!
!
! !
!
! !
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
! !
!
!
!
!
! !
!
!
! !
!
! !
!
!
0 !
! !
! !
!
!
! !
!
!
!
!
!
!
! !
!
! !
!
!
!
! ! ! !
!
!
!
!
! !
! ! !
!
! !
!
!
! !
! !
! !
!
!
!
!
!
! !
!
! !
!
!
!
! ! !
!
!
! !
! ! !
! !
! !
! !
!
!
!
! !
! !
! !
!
! !
! !
!
!
! !
! !
! ! !
!
!
!
! !
! !
!
!
! !
!
! !
!
! ! !
! !
! !
!
!
!
! !
! !
! !
! !
!
! ! !
! ! !
!
! ! !
! !
! !
! !
! !
!
!
!
!
! ! !
!
!
!
!
! !
!
! !
!
! ! !
!
! !
! !
! !
! !
!
! ! ! !
! !
! !
!!
! !
!
! !
!
!
!
!
!
! ! !
! !
! !
! !
!
!
!
0.660 !
!
! !
! ! !
!
! !
! !! !
!
! !
!
! !
! !
! ! !
! !
! !
! !
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
! !
!
!
!
!
!
!
!
!
!
! !
!
! ! !
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
!
! !
! !
! !
!
!
! ! !
!
!
! ! !
!
!
!
! !
!
!
!
!
! !
! !
!
! !
! ! !
!
! !
! !
! !
! !
!
! ! !
!
! !
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
!
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
! !
!
! !
! !
!
!
!
!
!
!
! ! !
! !
!
! !
!
!
!
! !
! ! !
!
!
! ! !
!
!
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! !
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! ! !
!
!
!
! !
!
!
!
! !
! !
!
!
!
2.953
! !
!
!
!
!
!
! !
!
! ! ! ! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
! !
! !
!
!
!
!
!
!
! !
! ! ! !
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! ! !
!
!
!
!
!
!
!
! !
!
!
!
!
! ! ! ! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
! !
! !
!
!
!
!
! !
!
!
!
! ! !
!
!
!
! !
!
!
!
!
!
! !
!
! ! ! ! ! ! ! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
! !
!
! !
!
! !
!
!
! !
!
!
!
!
!
!
! !
!!
!
!
!
! ! !
!
!
!
!
!
!
!
! !
! ! !
!
! !
!
!
! ! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
!
! !
! !
!
!
!
!
!
!
!
! !
! !
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
! !
!
! !
!
!
!
!
!
! !
!
!
!
! !
!
!
! !
!! ! !
!
!
!
!
!
!
!
! !
!
! !
! !
! ! !
!
!
!
!
!
!
!
! ! !
!
! !
!
!
! !
!
!
!
! !
!
! !
! ! !
!
! !
!
! !
!
! !
!
!
!
!
!
!
! !
! !
!
!
!
!
! ! !!
!
!
!
!
! !
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
! !
!
!
!
!
! !
!
!
! ! !
!
!
!
!
! !
! !
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
! !
! !
!
!
!
!
!
!
! !
! !
! !
!
!
!
!
! ! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!! !
!
!
! !
! ! !
!
! !
!
!
!
!
! ! !
!
! !
!
!
!
! !
! !
!
! !
!
!
! ! !
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
Radius: 600m !! Weight: count Beta: 0.004 Alpha: 1 !
!
! !
! !
! !
!
!
!
! !
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! !
! !
!
!
!
!
!
!
!
!
!
! !
!
! ! ! ! !
!
!
!
!
!
!
!
!
! ! !
!
!
! !
! !
!
!!
! !
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
! ! !
!
!
!
!
!
!
!
!!
!
!
! ! ! !
!
!
! !
!
!
!
D estinations: subway/bus stops !
!
!
! !
!
!
!
!
! !
!
!! !
!
! !
! !
!
! !
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
! ! !
!
!
!
! ! ! !
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
! !
!
! !
! !
!
!
!
!
!
!
! !
! !
!
!
!
! !
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
! ! !
!
! !
!
!
!
! !
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
! !
!
!
! !
! !
!
!
!
!
! ! !
! !
!
!
! !
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
!!
!
!
!
!
! !
! ! ! !
! !
!
! !
! !
!
! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
! !
!
!
!
!
! ! !
!
! !
! !
! !
!
!
! !
! !
!
!
!
! !
! !
!
!
! ! !
O rigins: all buildings
!
!!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! ! !
!
!
! !
! !
!
!
!
! !
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! ! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! !
! ! !
! !
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! ! !
!
!
! !
!
! !
!
!
!
! !
!
!
!
! !
! !
!
!
!
!
!
!
! !
! ! ! !
! ! ! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
! !
! ! !
! !
!
! ! !
! ! !
!
!
! !
!
!
!
!
!
! !
! !
!
!
! !
!
!
!
!
!
!
!
! !
! !
!
!
! ! !
!
!
! !
!
!
!
!
! !
!
!
!
! ! ! !
!
!
!
!
!
!
!
!
!
! ! !
! !
!
!
!
! !
! !
!
! !
!
! !!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! ! ! !
!
!
!
!
!
!
!
! !!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
! !
! !
!
! ! !
!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
! !
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!! !!
!
!
!
!
!
!
!
!
!
!
! !
! !
! ! ! ! !
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
! !
!!
! ! ! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
! ! !
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
! !
! !
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! ! ! !
! !
!
! !
!!
!
! !
!
!
!
!
!
!
! !
!
! !
! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
! !
!
! !
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! !
!
!
!
!
!
!
!
!
! !
!
!
! ! !
! !
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
! !
!
! ! !
!
! !
!
!
! !
!
!
! !
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
! ! !! ! ! !
!
!
!
! !
!
! !
! ! ! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
! !
!
!
! !
!
!
!
!
!
!
! ! !
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
! !
!
!
! !
!
!
! !
!
!
!
! !
!
!
! !
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! ! !
!
! !
!
!
!
!
!
!
!
! !
! !
!
! !
!
!
!
! ! !
!
! !
!
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
! ! !
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! ! !!
!
! !
!
!
! !
! !
!
!
!
!
!
!
! ! !
!
! !
!
! !
! !
!
!
!
!
!
!
!
! !
!
! !
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
!
!
! !! !! !
!
!
!
!
!
!
!
!
! ! !
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
! !
!
!
!
! ! !
!
! !
!
! ! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! ! !
!
! !
!
!
!
!
!
!
! !
! !
! !
! ! !
!
! !
!
! !
!
!
!
!
! !
!
!
!
!
!
!
! !
!
! ! !
!
! ! ! !
!
!
!
!
!
!
!
!!
!
!
!
!
! !
! !
!
! ! !
! !
! ! !
!
! ! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
! !
! !
! !
! !
!
!
!
! !
!
! ! !
! ! !
!
!
!
! ! ! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
! !
!
! !
!
!
!
! !
!
!
! !
!
! !
!
! !
!
!
!
!
!
! !
! ! !
!
!
!
! !
!
!
! !
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
! !
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
! ! ! !
! !
!
!
!
!
!
!
!
!
!
! !
! !
!
!
! ! !
!
!
!
!
!
!
!
!
! !
!
!
!
!!
!
!
!
!
!
! !
!! !!
!
!
!
!
!
! !
!
!
!
!
! ! !
!
! ! !
!
! !
!
!
!
! !
!
!
!
!
! !
!
!
! !
!
! !
!
!
!
!
! ! !
!
!
!
! !
! !
!
!
!
!
! !
! !
!
! !
!
! !
! !
! !
! !
!
!
!
!!
!
!
!
!
!
! !
!
! ! !
!
!
!
!
!
! !
!
!
!
! !
! !
!
!
!
!
!
!
!
!
! !
!
!
! ! !
!
! !
! !! !
!
!
!
!
!
!
!
!
!
!
! ! !
!
! !
! ! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
! !
!
!
! !
!
! !
!
!
!
! !
! !
!
!
!
! !
!!
! !
! !
!
! ! !
!
! ! ! !
!!
!
!
!
!
!
! !
!
!
!
! !
!
! !
!
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
! ! ! !
!
!
! !
! ! !
!
! !
!
! !
!! !
! ! !
! !
!
!
! ! !
! !
!
! !
! !
! ! !
!
! !
!
! !
!
!
!
! !
!
!
!
! !
!
!
!
! !
!
!
!
!
!
! !
!
!
!! ! ! ! !
Figure 62: This figure measures Gravity accessibility from each building to bus stops and ! subway!stations ! ! within a 600m network distance. Similarly, the Gravity accessibility values are smaller than the actual number of bus stops and subway stations that are within the desired radius due to the “distance! decay effect”. ! !
!! ! !
!
!
!
!
!!
! !
!
!
!
! !
! !
!
! !
! !
!! !
!! !!
!
!
! ! ! ! !
! !
! !
!
!!
!!
! !
!
!
!
!
!
! !
!
!
!
! !
!
!
! ! !
! !
!
!
! ! !
! ! !
! !
!
! ! !
!
!!
! !
!
!
!
!
!
!
!
! ! !
!
!!
! ! !
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
! !
!
!
! !
! !
!
!
!! ! !!
!!
! !
!
! !
! ! !!
!
!
!
!
!
!
!
!
!
!
! ! !
!
! !
!!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
89
!
!
!
bility Indices tool. Beta and the corresponding shape of distance decay should be derived from the assumed mode of travel and Destination type — for walking to retailers measured in “minutes”, for instance, researchers have found β to fall around 0.1813 (Handy and Niemeier 1997). This corresponds to a beta value of “0.00217” in meters. In moderate climates, a beta value of “0.002” is commonly used for many types of pedestrian Destinations. In tropical climates, such as Singapore, on the other hand, it is more typical to observe beta values around “0.004” (measured in meters) for walking trips. A higher beta value denotes a higher sensibility to walking distance. Since the beta values essentially approximate the likelihood to walk to particular Destinations – trips elasticities
!
90
analysis tools
with respect to distance — they also depend on Destination types. Many people are willing to undertake a longer walk to a subway station than to find a trash bin, for instance. See Sevtsuk (2018) on how a beta value of "0.001" (units = meters) was empirically determined for walking trips to subway stations in Cambridge, MA. Note that due to the negative effect of distance, Gravity values are always smaller than (or in exceptional cases equal to) Reach values, using the same Origins, Destinations and weights. Only in rare cases, where Origin points are found at the same locations as their Destinations, and where travel distances are effectively zero, can Gravity values be equal to Reach values. Figure 61 illustrates Gravity accessibility results from two bus stops (shown in blue) to surrounding building Destinations (shown in red) in a 200m Search Radius. Note how the resulting values are lower than the number of buildings that can be reached from both stops in the same walking radius due to the distance decay effect in the denominator of the index. Figure 62 shows a different set up, where buildings are used as Origins and public transit stations as Destinations. The results indicate accessibility to transit from each building. 4 . 4 .1 . 3 .
Equation 3
Straightness
The Straightness index (Vragovic, Louis, et al. 2005) illustrates the extent to which the shortest paths from Origins to Destinations resemble straight lines. Put alternatively, the Straightness metric captures the positive deviations in travel distances that result from the geometric constraints of the network in comparison to straight-line distances in a featureless plane. Straightness returns a number with a prefix “s” on each Origin point. The Straightness measure is formally defined as: Straightness[i]r =
j∈G −{i},d[i,j]≤r
δ[i, j] · W [j] d[i, j]
where Straightness[i]r is the Straightness between an Origin and a set of Destination points within Search Radius r, δ[i,j] is the straight-line Euclidian distance between i and j, and d [i,j] is the shortest network distance between the same points. Naturally, as the distances between nodes get longer, the proportional differences between the network distance and as-a-crow-flies distance start diminishing — a walk from Los Angeles to New York is much
URBAN NE T WORK ANALYSIS
91 Figure 63: The building on the left and the buildings within a 100m network distance from it are mostly situated along the same street. This building has a higher Straightness value than the building on the right, which is situated at an intersection.
s: 18.537
s: 30.591 radius: 100m
O rigins: buildings in blue ! !
!
!
!
!
!!
!
!
!
!
!
!
! !
!
! ! !
! !
!
!
! !
!
!
! ! !
!
!
!
!
!
! ! !
!
! !
!
!
!
!
!
!
!
!
! !
! ! !
! !
! !! !
!
!
!
!
!
!
! !
!
!
!
! !
!
!
! !
!
!
!
!
!
! ! ! ! ! !
! !
!
!
!
! !
! !
!
!
!
!
!
! !
!
!
!
!! !
!
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
! ! !
!
!
!
!
! !
!
!
!
!
!
!
! ! ! !
! !
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
! ! !
!
!
!
! !
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! ! ! ! !
! !
!
!
!
! !
!
! ! !
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
!
!
!
! !
!
! ! !
!
!
!
! !
!
! !
! ! !
!
!
! !
!
!
!
! !
! ! !
! ! !
!
!
! !
! !
! !
!
!
! ! !
!
!
!
!
!
! !
!
!
!
! !
!
! !
!
! !
! !
!
! !
!
! !
!
!
!
!
!
!
!
! !!
! !
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
!
!
!
! !
!
! !
!
!
!
!
!
! !
! !
!
! !
!
!
!
!
! !
!
!
!
!
!
! ! !
!
!
!
!
! !
! !
! ! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
! !
!
! !
! !
! !
! !
! !
!
! !
!
! ! !
! !
!
! ! !
!
!
!
!
!
!
! ! ! !
!
!
!
! !
!
!
!
! !
!
! !
!
!
!
!
!
!
! !
! !! ! !
!
!
! !
!
!
!
!!
! !
! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
! ! !
! !
! !
! !
!
!
!
!
!
!
! !
! !
!
! !
!
!
!
! !
!
!
! !
!
!
!
! !
!
!
! ! !
!
! !
!
!
!
!
!
!
!
! !
!
!
! !
! !
! !
!
!
!
! !
! ! !
!
! !
!
!
!
!
! !
! !
!
!
! !
!
!
! !
!
! !
!
!
! !
! !
! ! !
!
!
!
!
! ! !
! !
!
!
! !
! ! !
!
! !
!
!
!
! !
!
! !
!
! !
!
! !
! !
!
!
! ! !
!
!
! !
!
! !
!
! !
! ! !
!
! !
! !
!
!
! ! ! !
!
!
!
!
! !
! !
!
!
! !
!
! !
!
!
!
! ! !
! !
! !
!
!
!
!
! !
! ! !
! !
!
!
!
!
!
!
! !
! !
!
! !
!
!
!
!
!
!
! !
! !
! !
! !
!
!
!
!
!
! !
!
!
! ! ! ! ! !
! ! !
!
! !
! ! !
! !
! ! !
!
!
!
Straightness to buildings
!
!
!
!
! !
! !
!
!
!
! !
!
!
!
! !
!
! ! !
!
! ! !
!
!
!
! ! !
! ! !
!
!
!
!
!
! !
!
! !
!
!
!
! !
! ! ! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
! !
!
! ! !
! ! ! !
!
!
0
!
!
!
!
!
! !
! !
!
!
!
! !
!
! !
!
! ! !
! !
!
!
! !
! !
! !
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
! !
!
! !
!
!
! ! ! !
!
!
!
! !
!
!
! ! ! ! ! ! ! !
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
! ! ! ! !
!
!
!
!
!
! !
! !
! ! !
!
!
! !
! !
!
!
!
!
! !
! !
!
!
!
! !
! !
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
! ! !
!
! !
! !
!
!
!
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
!
! ! ! ! !
!
!
!! !
!
!
!
!
!
!
!
!
! !
!
!
!
! !
0.500
!
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
! ! !
! !
!
!
!
!
!
! !
!
!
!
!
! !
! !
! !
!
!
!
! !
!
! !
!
! ! !
! !
!
!
!
!
!
! ! !
!
!
!!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
! !
!
!
! !
! !
! !
!
!
! !
! !
! !
!
!
!
! ! !
! !
!
!
!
! !! ! !
!
!
!
! !
!
!
!
!
!
! !
!
!
! ! ! !
!
! !
!
! ! !
!
! !
!
! ! ! !
!
!
!
!
!
!
! !
!
!
!
!
! !
! ! !
!
!
!
! ! ! !
!
!
! !
!
!
!
!
! !
! !
! !
!
! !
! !
!
!
!
!
!
! !
! !
!
!
!
!
!
! !
!
!
! !
! !
!
!
!
!
!
!
! ! !
0.110
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
! !
!
! !
! ! !
!
!
!
!
!
! !
! !
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! ! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
! !
!
! !
!
!
! !
! !
!
!
! !
!
!
! !
!!
!
!
!
!
! ! ! !
!
!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
! !
! !
!
! !
!
!
!
! !
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
! ! !
! !
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
Radius: 600m Weight: count
! !
! !
!
!
! !
!
! !
!
!
!
!
!
!
! !
! !
!
! !
!
!
! !
!
!
! !
!! !
!
!!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
!
!
!
! !
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
!
!
! ! !
! !
!
! !
! !
!
!
! !
!
!
!
! !
! ! !
!
!
!
!
! !
!
!
!
!
!
!
!
! !
! ! ! ! !
!
!
!!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
! !
!
!
!
! !
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
!
!
! !
!
!
! !
!
! ! !
!
!
!
!
!
!
!
! !
!
! !
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
D estinations: all buildings
!
! ! !
! ! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
!
! !
!
!
! ! !
! !
! !
!
! !
! !
!
!
!
!
! !
! !
! !
! !
!
!
!
!
!
! !
! !
! !
!
!
! !
! ! !
! !
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
! !
!!
!
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
!
!
!
! !
!
!
! ! !
!
O rigins: all buildings
!
! !
!
!
!
!
!
!
! !
!
! ! !
!
!
!
!
!
!
!
!
!
!
! !
! ! ! ! !
!
! !
!
!
!
!
!
!
!
!
! ! ! !
! !
!
!
!
! !
! ! ! ! !
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
! ! !
! !
!
!
! !
! ! !
!
! !
!
!
!
!
! !
!
!
!
!
!
!
!
! ! !
!
!
! !
!
! !
!
!
!
!
! !
! !
!
!
!
! ! !
!
!
! ! !
! ! !
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
! !
!
! !
!
!
! !
!
! ! !
! !
! !
!
!
! !
!
!
!
!
!
! !!
!
!
!
! !
! !
!
!
!
!
! !
! !
! !
!
! !
! !
! !
!
!
! ! !
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! ! ! !
! ! ! !
!
! ! !
!
!
!
! !
! !
!
! ! !
!
!
!
!
!
! !
!
!
!
! !
!
! !
!
! !
!
!
!
!
!
!
!
! !
! !
!
!!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
!
! ! !
! ! !
!
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
! !
!
! !!
! ! ! !
!
!
! !
!
!
!
!! !!
!
!
!
!
!
! !
! !
!
!
!
! ! !
!
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
! ! ! ! !
! !
!
!
!
!
!
! ! ! ! !
!
!
! !
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
!
! !
! !
!
!
!
!
!
! !
! !
!
!
!
!
! !
! !
!
!
!
!
! ! !
! !
!
!
! !
! ! !
!
! !
!
! !
!
! !
!
!
!
! ! !
!
!
!
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! ! !
! !
! !
!
!
!
!
! !
! ! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
! !
!
!
!
!
!
! ! !
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
! !
!
! !
! !
!
! !
!
! !
!
! !
! ! !
!
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
! ! !
!
! ! !
!
!
!
! !
!
!
!
! ! !
! !
!
!
!
!
!
!
! !
!
!
! !
!
! ! ! ! !
! !
!
!
!
!
!
!
!
! ! !
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
! !
!
! !
!
!
!
! !
!
!
!
! !
!
!
! !
!
!
!
!
!
!
!
! !
! !
! !
!
!
!
! !
!
!
!
!
!
! !
!
! ! ! !
!
!
!
!
! !
!
!
! !
!
!
! !
!
!
!
!
!
!
!
! !
!
! !
! ! !
! !!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
!
!
!
! ! !
! !
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! !
!
!
!
!
! ! !! ! ! !
! !
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
! ! ! ! !!
!
! !
!
!
!
!
!
! !
! !
! !
!
!
!
!
! !
!
!
!
! !
!
!
!
!
!
!
!
! ! !
!
!
! !
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
!
!
! ! !
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
! !
!
!
!
!
! !
!
!
! !
! !
! !
!
! ! !
!
! !
!
! !
!
!
!
! ! !
! !
!
!
! !! !! !
!
!
! !
! !
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
! !
!
Figure 64: Buildings who can reach most destinations along straight travel paths obtain highest Straightness values. Low Straightness values may indicate places that require complex routes to reach. !
!
!
! !
!
!
!
!
! !
! !
!
!
! !
!
! ! !
!
! !
!
!
!
!
!
! ! !
!
!
!
! !
!
! !
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
! !
!
!
!
! !
!
! !
!
! ! ! !
!
!
! !
! !
!
!
! !
!
!
!
! !
! !
!
!
! ! !
!
!
!
! !
!
! ! !
! !
! ! !
!
!
!
! !
! !
!
! !
!
! !
! ! ! !
!
! !
!
!
! !
! !
!
!
! !
!
! ! !
!
! !
!
!
! !
!
!
! ! !
!
!
! !
!
! !
!
!
!
! !
!
!
! ! !
!
! !
!
!
!
!
!
! ! !
!
!
!
!
! !
!
!
!
!
!
!
!
!
! !
! ! ! ! !
!
!
!
! ! !
! !
!
!
!
!
! !
! !
! !
!
!
! !
!
!
!
!
!
!
!
! !
!
!
! !
!
!
!
!
!
! !
!
! ! !
!
!
!
!
!
!
! ! !
! !
!
!
! !
!
!
!
!
! !
!
!
!
! !
! !
! !
!
!
!
! !
!
!
!
!
!
!
! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! !
!
! ! !
!! !
! !
!
!
!
!
! ! !
!
!
!
!
!
!
!
! ! !
!
!
!
!
!
! !
!
! !
!
! !
!
! !
!
!
!
!
!
!
!
! ! !
!
! !
! ! !
!
!
!
!
!
!
!
!
!
!
!
!
! ! !
! !
! !
!
!
! ! ! !
!
!
!
!
!
!
!
!
!
!
! !
!
!
! ! ! !
! !! !
! !
!
!
! !
! !
! !
!
!
!
! !
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
! !
!
! !
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
! !
! !
!
! !
! !
!
!
!
! ! !
! !
! !
!
!
!
! !
! !
! !
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
! !
! ! !
!
!
! !
!
!
! ! !
!
!
!
!
!
!
!
! !
! !
!
!
!
! !
!
! !
! !
!
!
!
! !
!
!
! !
!
!
!
!
! !
! !
!
!
!
!
! ! ! !
!
!
Radius: 100m Weight: count
! !
!
!
!
!
! !
!
!
! !
!
!
!
!
! ! !!
! !
!
!
!
!
!
! !
! !
!
!
!
!
! !
!
! !
!
! !
! ! !
! !
! ! ! !
!
!
! !
!
! !
! ! !
!
!
!
! !
!
!
!
!
!
!
! ! !
!
!
!
! ! !
!
!
!
!
!
! !
!
! ! !
!
!
! !
!! !
! ! !
! !
!
!
!
!
!
! !
!
!
! !
!
!
!
!
! !
! !
!
! !
!! ! ! ! !
! !
!
!
!
! !
!! ! !
!
!
!
D estinations: all buildings
!! !
!! !!
!
!
!
!
!
!
!
!
! !
! !
!
!!
!!
! !
!
!
!
!
!
!
! ! ! !
!
!
!!
! ! !
! !
!
! !
!
! !
!
!
!
!
!
!
!
!!
! !
! !
!
! !!
!!
!
!
! !
!!
!!
! !!
! !
! !
!
!
!
!
! !
! !
! !
! !
!
!
!
92
analysis tools
closer to a straight line than a walk from downtown LA to Culver City. This bias should be kept in mind when interpreting Straightness results. The fact that Straightness illustrates how much longer actual network walks from Origins to Destinations are in comparison to as-a-crow-flies distances, makes it a handy metric detecting points of frustration in a city, where Destinations may appear seemingly close, perhaps even visible, but are actually inaccessible due to long travel detours. The index can be used, for instance, to predict frustration points for pedestrians, where travel Destinations, such as bus stops or building entrances are visually near, yet the routes to reach them involve considerable detours. With single destinations, results vary from zero to one and they are intuitive to interpret â&#x20AC;&#x201C; a Straightness value of 0.75 indicates that "as a crow flies" distance to the destination constitutes 75% of the shortest available network route to the same destination. This is equivalent to saying that the network route requires a 25% detour compared to the line of sight. With more destinations, such intuitive interpretations dissapear, since several Straightness values are summed. To make analyses that involve multiple destinations comparable, users can divide Straightness results by Reach values for the same Origin-Destination set and Search Radius, which produces average Straightness results. Figure 64 illustrates Straightness results from all buildings to all other buildings in Cambridge, MA in a 600m Search Radius. Origins, which have more Destinations close by or that are surrounded by a denser street pattern, tend to reach Destinations along more direct paths. 4 . 4 . 2. SERVICE Figure 65: Service Area tool
AREA
The Service Area tool selects or copies Destination points and path segments that fall within a given Search Radius from Origins. The tool can be used, for instance, to select all restaurants (Destinations) that fall within a 200m network radius from a set of bus stops (Origins). The Search Radius has to be typed into the command line. The tool offers four options: Service area <600> (SelectPoints=On SelectCurves=On Copy=Off Tight=Off):
URBAN NE T WORK ANALYSIS
93 Figure 66: This figure presents the same results as in the example for the Reach tool. If the option “ Tight” is switched off, all curves that are connected to the Destination points within the radius buffer will be selected regardless of whether the curves themselves are within the radius. Origins: building in blue Destinations: all buildings Radius: 100m Selectpoints: on Selectcurves: on Tight: off
Figure 67: This figure shows the selection of curves and Destination points that are within a 600m network distance from subway stations. Origins: subway stations Destinations: all buildings Radius: 600m Selectpoints: on Selectcurves: on Tight: off
94
analysis tools
SelectPoints = On and SelectCurves = On options enable the analysis to select the Destination points or curves that fall within the specified Search Radius (Figure 66). You can turn point selection on and curve selection off, for instance, to only select Destination points. Note that curves are selected with their full original lengths and not cut into shorter curves that exactly correspond to the Service Area Search Radius. Copy = Off option enables you to copy the selected points and curves to the active layer. Make sure to check what layer is currently active or activate another layer to copy the points and lines elsewhere. This option is turned off by default. Tight = Off option, which only affects curve selections, allows the tool to only select curves that are fully enclosed within the Search Radius buffer. Figure 67 illustrates how the Service Area tool selects buildings (Destinations) and segments that are within a 600m Search Radius around a subway (Origin) in Cambridge, MA. 4 . 4 . 3 . REDUNDANCY Figure 68: Redundancy Index tool
INDEX
The Redundancy Index computes the increase in linear path distance that becomes available when the shortest walk between an Origin and Destination is extended by a given percentage, called Detour Ratio (Sevtsuk et al. 2014 a, b). This can help describe the extent of reasonable alternative travel paths to given Destinations around a set of Origins. Many urbanists have argued that having more path options to Destinations is a positive quality of the built environment enabling travellers more choices. If you undertake a routine walk every day — say from your house to the nearest bus stop — the index can help you estimate how many other streets you could use within a given Detour Ratio, and by extension, what Destinations you could potentially visit around your everyday route, without much of a detour. Redundancy Index returns a number with a prefix “ri” on each Origin point. It also returns a Route Count "rc" results, indicating how many unique routes to destinations are found with the given Detour Ratio. Unique routes differ by at least on segment and backtracking or repetition of nodes is not allowed. The tool first finds alternative redundant routes between an Origin and Destinations and when all available routes are found, the index is returned as a ratio between the combined lengths of all routes between an Origins and Destinations divided by the length
URBAN NE T WORK ANALYSIS
ri: 6.42 rc: 56
of the shortest paths between the same Origin and Destinations. The result is interpreted as a factor that sizes how much more linear street length becomes available when the shortest walk is extended by a given percentage. This percentage input is called a Detour Ratio, and constrained in the UNA tool between one and two. A Detour Ratio “1.2” means that all routes that are up to 20% longer than the shortest path will be analyzed. Given a pair of nodes i and j in a positively weighted, undirected graph G, the Redundancy Index R for the node pair with a Detour Ratio ρ ≥ 1 is defined by:
W [e] · ζ i,j [e, ρ · d[i, j]] R [i, j] = e∈G i,j e∈G W [e] · ζ [e · d[i, j]]
95
Figure 69: The black route denotes the short route to travel from the building in blue to the building in red. Given a Detour Ratio of 1.2, which considers all possible routes that are at most 20% longer than the shortest route, the Redundancy Index (ri) of 6.42 indicates that the total length of routes expands 6.42 times compared to the shortest walk. The Redundancy Count (rc) indicates that there are 56 routes in total that are at most 20% longer than the shortest route. Origins: building in blue Destinations: building in red Detour ratio: 1.2
ρ
where W [e ] is the weight of Observer points on an edge e in G; d [i,j ] is the shortest path distance from i to j in G; and ζ i,j [e] is 1 if there exists a simple path from i to j in G that goes through edge e and 0 otherwise. A simple path is a path where no nodes are repeated — that is, no backtracking or looping is allowed, though there are some exceptions, where the routes may contain loops and retrace repeat nodes as described in Sevtsuk et al. (2014). The enumerator of the index sums the lengths of all Redundant Path segments between the Origin and Destination; the denominator sums the lengths or weights of the shortest path segments between the same Origin and Destination. When the index is weighted with Observer points, the results illustrate how many more Observer points become accessible on a walk from an Origin to a Destination along Redundant Paths
Equation 4
96
analysis tools
Figure 70: The analysis measures Redundancy Indices for all buildings with subway stations as Destinations. Intuitively, the buildings with higher Redundancy Indices are the ones further away from the subway stations, because there are more routes to choose from that are at most 20% longer than the shortest route. For the buildings close to the subway stations, there is often only one way to walk there, which explains the Redundancy Index of 1. Origins: all buildings Destinations: subway stations Detour ratio: 1.2 11.90 5.95 1 Redundancy index to subway stations
compared to the shortest path alone. This can help describe how many more address points, coffee shops, trees or whatever other route features can be passed by reasonably extending the walk. When the user inputs more than one Destination, then the Search = Nearest option on the command line tells the algorithm whether to search for only the Nearest Destination for each Origin, All Destinations for each Origin or only Destinations that fall within a given Radius from the Origin (e.g. 200m). Draw = On option controls whether all routes that are found to satisfy the Detour Ratio are copied and grouped as a new Rhino object onto the same layer as the original routes for later manipulation.
URBAN NE T WORK ANALYSIS
97
The resulting Redundancy Index “ri ” is equal to or greater than one. If it is one, then no Redundant Paths were found within a given Detour Ratio. If it is “5.5”, for example, then the cumulative length of the Redundant Paths was 5.5 times longer than the shortest path between the given Origin and Destination. Besides the Redundancy Index “ri”, the tool also outputs the redundant Route Count “rc”, that counts the number of alternative routes that was found. Note that routes, where even a single street segment differs, are counted as separate. The same Route Count “rc” is also output by the Betweenness tool if a Detour Ratio is applied there. Figure 70 illustrates the Redundancy Index from all buildings in Cambridge, MA to a subway station using a Detour Ratio of “1.2”. This means that all routes that are up to 20% longer than the shortest path are found from each building to he subway station and the resulting index measures how much more street length becomes available as a result of the detours. Naturally, buildings that are further away from subways obtain a higher outcome — they have more routes to choose from to go to the subway, which are at most 20% longer than the shortest. For buildings that are close to subway stations, there is often only one path that satisfies the detour constraint, which is why many of them obtain a Redundancy Index of “1”. 4 . 4 . 4 . REDUNDANT
PATHS
The Redundant Paths tool finds all individual paths between a set of Origins and Destinations that are up to a given percentage — defined by the Detour Ratio –longer than the shortest path and outputs those paths as new polyline objects. The tool can be used to study pedestrian route choice. It allows one to identify all plausible paths that a person might be expected to walk between an Origin — Destination pair. Unlike the Redundancy Index, the paths found by the Redundant Paths tool are truly simple paths — they contain no loops or repeating nodes. The tool outputs polylines, where each polyline is a separate path. The polylines are not grouped — each individual route is drawn separately and precisely from where an Origin enters and a Destination exits the network. The polylines are placed on the active Rhino layer. Note that depending on the connectivity of the path network and the relative position of Origins and Destinations, the number of Redundant Paths can be very high, slowing down Rhino operations.
Figure 71: Redundant Paths tool
98
analysis tools
Figure 72: This figure shows the selection of all routes that are at most 20% longer than the shortest route between the building in blue and the building in red.
ri: 6.42 rc: 56
O rigins: building in blue D estinations: building in red Detour ratio: 1.2
Figure 73: This figure shows the selection of all routes that are at most 5% longer than the shortest route between the Harvard T station and the Kendall T station. Note that the number of redundant routes as well as the computing power required for this analysis increase exponentially with the Detour Ratio. In this example, a Detour Ratio as low as 1.05 already yields 475 routes. O rigins: Harvard TÂ station D estinations: Kendall TÂ station Detour ratio: 1.05
ri: 6.60 rc: 475
URBAN NE T WORK ANALYSIS
99
The tool offers several options: Search = Nearest option determines whether to search for only the Nearest Destination for each Origin, All Destinations for each Origin or only Destinations that fall within a given Radius from the Origin (e.g. 200m). Detour Ratio = 1 option defines the proportion of allowable detours — the length of paths to look for, compared to the shortest path. The ratio is constrained between one and two. Using a Detour Ratio of “1.3”, for instance, tells the algorithm to find all paths that are up to 30% longer than shortest paths. Start = Network option determines whether the polylines that the tool outputs start at the network locations of Origins and Destinations (i.e. where Origins and Destinations snap to the network) or at the actual locations of Origins and Destinations (e.g. exactly at the Origin and Destination points, possibly outside of the network). Draw = On option controls whether all routes that are found to satisfy the Detour Ratio are copied and grouped as a new Rhino object onto the same layer as the original routes for later manipulation. Figures 72 and 73 illustrate Redundant Paths (red curves) between Origins and Destination that satisfy a Detour Ratio of “1.2” and "1.05" respectively. In both examples, using a detour allows pedestrians to access six times more cumulative street length (and potentially building frontage) compared to the shortest path length. 4 . 4 . 5 . BETWEENNESS
The Betweenness tool in the UNA toolbox calculates and visualizes how many trips are likely to pass different network edges or Observer points, given a set of trips between Origins and Destinations in a network. The tool is often used to approximates foot or bike traffic at particular locations or across an entire network. If applied within buildings, the Betweenness index can also estimate passerby at different points within a three-dimensional circulation structure. Originally proposed by Freeman (1977), the Betweenness centrality of a node in a network is defined as the fraction of shortest paths between pairs of Origins and Destinations that pass by a particular location. If more than one shortest path is found between two nodes, as is frequently the case in a rectangular grid of streets, for instance, then each of the equidistant paths is given
Figure 74: Betweenness tool
100
analysis tools
Equation 5
equal weight such that the weights sum to unity. The measure is mathematically defined as follows:
Betweenness[i]r,dr =
P
j,k2G −{i},d[j,k]r·dr
nj,k [i] nj,k
· W [j] ·
1 e
·d[j,k]
Where Betweenness[i ]r,dr is the Betweenness of an Observer location i within Search Radius r, and Detour Ratio dr, nj,k[i ] is the number of shortest paths from Origin j to Destination k that pass by i, and nj,k is the total number of shortest paths from j to k. Betweenness for a location i is computed by considering all pairs of Origins and Destinations that are within a distance r from each other. Therefore, distance r refers to a trip distance from Origin j to Destination k, not the distance from Observer point i to either Origin j or Destination k. When a Detour Ratio greater than one is used, individual routes between j and k can be longer than r. The last term in Equation 6 (1/e β ·d [j,k ]) is only applied when the Gravity option is turned on, as discussed below. When the Betweenness measure is weighted, the weights W[j ] are attached to the trip Origin points j. If an Origin point carries a weight of “100”, for example, then 100 trips are routed from that Origin to its Destinations. Just like with the Redundancy Index or the Redundant Paths tools above, the reliance on shortest paths can be relaxed using a Detour Ratio variable. Detour Ratio refers to a ratio between acceptable path lengths and the shortest path lengths. A Detour Ratio “1.2” means that all routes that are up to 20% longer than the shortest path are included in the analysis. The UNA Betweenness tool assumes that all paths that satisfy the given Detour Ratio are equally probable. If, for instance, using a Detour Ratio of “1.2”, finds 27 different paths from an Origin to a Destination, and no weights are used, then 1/27 or approximately 0.03704 trips are routed along each of the 27 paths. If the measure is weighted and the Origin has a weight of “100”, then each of the 27 paths receives 100/27 or approximately 3.704 trips. On the network segments, where the 27 paths overlap, Betweenness results are cumulatively summed (Figure 75). When running the Betweenness tool, the command line first prompts you to validate Origin points, Observer points and Destination points. Observer points are optional, while both Origin and Destination points are required. The Observer points do not emit or receive any trips themselves, they can be simply used to count how many trips pass by specific locations.
1
URBAN NE T WORK ANALYSIS
The tool presents the following options: Betweenness (Search=Nearest DetourRatio=1.2 Weight=Count Gravity=On Beta=0.002): Search = Nearest option determines whether to search for only the Nearest Destination for each Origin, All Destinations for each Origin or only Destinations that fall within a given Radius from the Origin (e.g. 200m). Detour Ratio = 1 option defines the proportion of allowable detours — the length of paths to look for, compared to the shortest path. The ratio is constrained between one and two. Using a Detour Ratio of “1.3”, for instance, tells the algorithm to find all paths that are up to 30% longer than shortest paths. Weight = Count option allows you to weigh the Betweenness results with Origin weights. When estimating walking trips in a residential neighborhood, for instance, Origin weights could refer to the number of inhabitants that are expected to make trips from each building. When the weights are kept as Weight = Count, then each Origin point simply emits “1” trip and no additional weights are applied. You can choose weights by clicking on the Weight = Count option, which will list all the numeric attribute weights that are available in the drawing. Click on the weight field you want to use and make sure that the weights you assign are actually associated with the Origin points in your network. Gravity = Off option enables the Betweenness estimates to be adjusted with a distance decay effect, depending on how far trip Origins and Destinations are from each other. This enables the model to assume that the further away Origins and Destinations are from each other, the less likely we are to witness any trips between them. The distance decay effect is identical to the one used in the Gravity accessibility metric. Turning the Gravity option On, multiplies each Origin point’s weight W[j] by an inverse distance ratio using an exponent beta: W [ j ] · 1/e d[j ,k]· β. As a result of enabling the Gravity effect, Origin points emit fewer trips as the distance to Destinations increases. The effect is exponential, controlled by exponent beta, in the same way as in the Gravity metric above. Consider an example, where households are used as Origins and bus stops as Destinations. Shortest walking distance from a particular household that carries a weight “10” to its nearest bus stop is 500 meters. If Gravity = Off, then 10 trips are routed to
101
102
analysis tools
Figure 75: This analysis computes Betweenness values for both curves and points in the network. In this example, the Betweenness values are weighted by the number of residents at the Origin, which is 100. The values next to the paths indicate the number of residents from the Origin that would take them as they walk to the Destination, assuming people only take routes that are at most 20% longer than the shortest route. O rigins: building in blue D estinations: building in red Observer points: all buildings Detour ratio: 1.2 Weight: residents 100
60.7
0
Betweenness
35.7
35.7
28.6
ri: 6.42 rc: 56
21.4
21.4
7.1
60.7
64.3
69.6
100 12.5
28.6
21.4
7.1
26.8
19.6
21.4 33.9
28.6
5.4
26.8
17.9
3.6
16.1 21.4
32.1
39.3
25
19.6
12.5
3.6
12.5
5.4
100 37.5
17.9
7.1
1.8
3.6
8.9
35.7 1.8
35.7
26.8
8.9
12.5 10.7 14.3
12.5
8.9 10.7
12.5
8.9
8.9 5.4
10.7
5.4
the Destination along all acceptable routes. If Gravity = On, then less than 10 trips are sent out of this household. The exact number of trips sent out depends on the beta coefficient used. Beta determines how sensitive we assume travelers to be to increasing distance. With a beta value of “0.002”, the number of trips sent out would be 10 * 1/e^(0.002*500) = 3.68, instead of the original 10. But due to the exponential shape of the “distance decay” rate, the same beta value of “0.002” would penalize a household that is closer to the bus stop less. A household located only 65 meters from the same bus stop, which also houses 10 inhabitants, would send out 10 * 1/e^(0.002*65) = 8.78 trips. Turning on the Gravity effect always decreases the number of trips between Origins and Destinations, but the scale of the penalty depends on trip distance. Beta and the corresponding shape of “distance decay” should be derived from the assumed mode of travel and Destination type — for walking to retailers measured in “minutes”, for instance, researchers have found β to fall around 0.1813 (Handy and Niemeier 1997). This corresponds to a beta value of “0.00217” in meters. In moderate climates, a beta value of “0.002” is commonly used for many types of pedestrian Destinations. In tropical climates, such as Singapore, on the other hand, it is more typical to observe beta values around “0.004” (measured in meters) for walking trips. A higher beta value denotes a higher sensibility to walking distance. Since the beta values essentially approximate the likelihood to walk to particular Destinations – trips elasticities with respect to distance — they also depend on the Destination types. Many people are willing to undertake a longer walk to a subway station than to find a trash bin.
URBAN NE T WORK ANALYSIS
103 Figure 76: Setting subway stations as Destinations, the highest Betweenness values are found around the subway stations as well as main arterial roads that are connected to them. O rigins: all buildings D estinations: subway stations Observer points: all buildings Detour ratio: 1.2 Weight: residents 332.26
64.64
0.01
Betweenness
Figure 76 illustrates estimated footfall in front of each building and on each street segment in Cambridge, while trips are assumed to flow from homes to subway stations during morning peak hour. 4 . 4 .6 . CLOSEST
FACILITY
The Closest Facility tool identifies the closest Destination facility to each Origin point and summarizes Reach or Gravity values at each Destination facility. Unlike the Reach and Gravity indices discussed above, each Origin point here is used only once and
Figure 77: Closest Facility tool
104
analysis tools
Figure 78: This figure shows the buildings that are assigned to their respective closest bus stops within a 100m network distance. If there is more than one bus stop within the designated network distance, the one closest to the building in question will be assigned. Since the Gravity function is turned on, the “distance decay” effect is accounted for. As such, the Gravity numbers next to the bus stops are smaller than total number of residents they serve respectively.
18.84
51.93
37.18
57.30
O rigins: all buildings D estinations: bus stops in red
1016
Radius: 100m Weight: residents Gravity: on573 Beta: 0.004 Figure 79: The figure shows the assignment of buildings to their nearest fire station in Cambridge if there is any within a 1000m network distance. The Gravity function is off, and so the numbers next to the fire stations are equal to the total number of residents they serve respectively. O rigins: all buildings D estinations: fire stations Radius: 1000m Weight: residents Gravity: off
546 2100
1922
327
URBAN NE T WORK ANALYSIS
summarized at the facility that it lies closest to (no Origins are counted multiple times, which may well occur in running Reach and Gravity metrics due to walkshed overlaps between adjacent Origins). The tool can optionally trace idealized straight lines from each Origin to its nearest Destination facility, visualizing the allocation of facilities between different Origins. The command line choices offer the following options: Search Radius <600> (Weight=Count Gravity=Off Beta=0.004 Lines=On NameAs): Search Radius determines the maximum network distance for linking Origin points to Destination facilities. Weight option allows you to weigh the analysis with Origin attributes. If an Origin has a weight of “15”, then it contributes “15” units to the Reach value at its closest Destination facility. When the weights are kept as Weight = Count, then each Origin point is simply counted as “1”. You can choose weights by clicking on the Weight = Count option, which will list all the numeric attribute weights that are available in the drawing. Click on the weight field you want to use and make sure that the weights you assign are actually associated with the Origin points in your network. Gravity = Off option determines whether or not Gravity results are computed for facilities. By default, only Reach values are computed. Beta = 0.004 option determines the beta value used in the Gravity accessibility metric. This is only used if the Gravity option is set as Gravity = On. Please see the Gravity accessibility section above for more details about interpreting the beta coefficient. Lines = On options allows the tools to draw a straight line from each Origin to its closest Destination facility. The lines are created on the active Rhino layer. Note that while these lines are straight for visual simplicity, Origins are actually related to closest Destination facilities along network distances. NameAs option allows you to assign a custom name for the output results. Choosing a name will save a new numeric attribute for each Destination point, its value corresponding to how many Origins found this Destination to be their Closest Facility.
105
106
analysis tools
4 . 4 .7. FIND
Figure 80: Find Patronage tool
PATRONAGE
The Find Patronage can be used to predict the patronage of spatial facilities (city parks, playgrounds, shops, libraries, bike-sharing facilities etc.) on networks, from given demand Origins in the presence of competing facilities. It assigns probabilities for trips from each Origin point to each Destination facility based on how accessible the destinations are. The tool is based on David Huff’s (1963) seminal patronage model as well as the subsequent work of Eppli and Shilling (1996). A paper by Sevtsuk and Kalvo (2017) describes an implementation of the tool on retail center planning in Singapore’s housing towns, which is also illustrated by a case study at the beginning of this document. The tool includes modifications that differ from the original Huff model (Sevtsuk and Kalvo 2017), which customize it for implementation in urban design and planning practice. The classical Huff model assumes that the probability of a consumer at a certain Origin to visit a given commercial Destination is a function of the Destination’s attractiveness, distance from the consumer, as well as the presence of competing Destinations around the consumer. The attractiveness attribute of each Destination facility can describe any measurable quality that has a positive effect on consumer patronage. In practice, area is often used as a proxy for the choice of offerings at each Destination, which is known to have a positive effect on patronage. But Destination attractiveness could also capture variations in retail prices, parking spaces, street frontage, expenditure on advertising and so on, all of which are typically combined into a single attractiveness index. Since a certain proportion of each consumer’s visits are allocated to each Destination, an interaction ensues between all consumers and all Destinations. From each consumer (Origin), a Gravity accessibility index is computed to each Destination facility that is found within a given Search Radius. This Gravity accessibility is proportional to the attractiveness of the Destination divided by the travel cost of getting there (see more about Gravity under Accessibility Indices above). The probability that the consumer will visit the particular center is found as a ratio between the Gravity accessibility to that center, divided by the sum of Gravity accessibilities to all centers, including the one under question. More attractive Destinations and closer Destinations obtain relatively higher probabilities. But as long as each Destination has non-zero attractiveness, no center
URBAN NE T WORK ANALYSIS
is left with a zero probability — even the most remote and poorly attractive facilities get some customers due to some randomness in choice making. The result combines these probabilities with Origin weights, estimating the number of visitors or their weights (e.g. disposable income) at each Destination facility. The tool is composed of two parts: a left click on the icon opens the Find Patronage tool, a right-click opens the Find Patronage Window tool. Both estimate patronage of facilities but the Find Patronage Window tool works with a graphic user interface and contains additional functionality that allows users to differentiate Destination facilities into up to three different categories, each of which can have a different “distance decay” coefficient beta and a different Search Radius for finding Destination facilities. This makes the Find Patronage Window tool suitable for hierarchical facility networks, such as primary, secondary and tertiary Destination types. For most applications, the simple Find Patronage tool with a command line interfaces suffices. Before using the tool, you need to have built a network (using Add Curves to Network tool) and to added both Origin and Destinations points to the network (using Add Origins and Add Destinations tools). Origins can optionally contain weights that indicate demand for facilities. For instance, if building entrances are used as Origins, then buildings’ weights could indicate the number of residents at each building or their income. Destination facilities can also contain a numeric weight that can be used to model their attractiveness. An “area” weight, for instance, can indicate the estimated sizes of retail Destinations, which can be accounted for in allocating demand to facilities. The Find Patronage command line options include the following: Search Radius <800> (OriginsWeights=Count DestinationWeights=Count Beta=0.004 Alpha=1 ApplyImpedance=On CopyResultsToMemory=Off): Search Radius defines the maximum distance between Origins and Destination facilities, beyond which no trips are made to visit facilities. In case of retail facilities, for instance, if a household is located further than the Search Radius from a particular store, then none of the household demand is allocated to that store. Note that Search Radius units follow the Rhino drawing units in your model.
107
108
analysis tools
Figure 81: This figure shows the number of residents visiting each of the three businesses, assuming they have equal attractiveness ( alpha=1). Note that even though the business on the far right is not closest to any of the buildings out of the three businesses, it still draws visitors. This is because the probability of visiting a business is defined as the ratio between the Gravity accessibility value of that business and the sum of all Gravity values, so that not all residents will necessarily visit the nearest business. O rigins: all buildings D estinations: businesses in red
Origin weights: residents
Dest’n weights: count
Beta: 0.004 Alpha: 1 Applyimpedance: on
54.248 residents 30.288 residents
51.678 residents
Origin Weights = Count defines the attribute weight of the Origin that is used to estimate demand (e.g. residents at a building). The default option “Count” makes the tool simply count the Origin points, without factoring in their weights. In this case, all demand Origin points are treated equally — they all send out “1” hypothetical trip, which is split among different Destination facilities. Destination Weights = Count defines the attribute weight of the Destination facilities, used to estimate their attractiveness (e.g. area, capacity, brand recognition etc.). The default option “Count” makes the tool apply a weight of “1” to each Destination. In this case, all Destination facilities are assumed to be equally attractive. Beta = 0.004 option defines the effect of “distance decay” used in allocating Origin weights to Destination facilities. Just as in the Gravity index (see above), beta ranges from 0 to 1. If beta is less than 1, then the amount of demand allocated to a facility decreases exponentially as the distance to the facility grows. Beta should be derived from the assumed mode of travel - for trips to retailers measured in “meters”, for instance, beta tends to vary between 0.0005 and 0.002, where a higher value denotes greater aversion to distance. In the context of Singapore, we have determined that beta for cconvenience retail walking trips averages around 0.001 (Sevtsuk and Kalvo 2017). If no value is provided by the user, a default value of “1” is assumed. See below for equations that use the beta coefficient. Alpha = 1 option defines the effect of Destination attractiveness on patronage. If Destination weights indicate store areas in square meters (e.g. 1,500 m²), then alpha defines how the effect
URBAN NE T WORK ANALYSIS
109
84
2312
145
17
145
17 17
6
3341
95
69 175
23
48
00 1
27
36
95 58
31 32 7 51 3 18 85 42 54
57
40 2831
40 88
62
14
29 5 3 27
82
124
11 5 43 1
Figure 82: This analysis computes patronage from all buildings to parks in Cambridge, weighting the Origins by number of residents and the Destinations by their square footage, which is a proxy for a park’s attractiveness. The numbers next to the parks indicate the number of residents who would visit each of them, adjusted for “distance decay”.
28
55
129 30
61 53
O rigins: all buildings Destinations: parks in red
102
45 8
20 9 38 24 25
9 Dest’n weights: 108 30park area 112
14 63
48
Origin weights: residents
Beta: 0.004 Alpha: 1 129 148 65 85 Applyimpedance: on 14 252 98 95 683 77 372 73
137
8
45
of store area affects attractiveness as the area changes. Alpha is modeled as an exponent to area (e.g. weight^α) as part of the Gravity accessibility component. See below for equations that use alpha. If no value is provided, a default value of “1” is assumed. ApplyImpedance = On option defines whether or not a “distance decay” function is applied to Origin points, depending on their distance from Destination facilities. If applied, then not all Origin demand weights reach the Destination facilities, resulting in a lower total patronage results across all Destinations than the sum of Origin demand weights would suggest. If a household is estimated to allocate 2 customers to a particular café located a mile away, for instance, then turning Apply Impedance = On and
78
74
116 105
26 66 137
0
120 131 139Patronage
33 50
73 27
59
177 64
149 60 20
64
22 10
683
110
analysis tools
Equation 6
using a beta coefficient for distance decay, would only allocate a fraction of the original “2” patrons to the café [2/e^(distance*beta)], since travel costs reduce patrons’ probabilities to visit cafés at a distance. Similar households located closer to the café, would allocate a larger proportion of their original demand to the café. This is similar to the Gravity function described above. See below for equations, where ApplyImpedance is turned On or Off. CopyResultsToMemory = Off creates a clipboard copy of each Destination facility’s estimated patronage results, which can be pasted to Excel or other table editing software for further analysis. The following explains the mathematical definitions of the different options in the tool: Let “DP” = Demand Point (Origin “i ”) and let “C ” = Center (Destination “j”). The Gravity accessibility from a demand Origin “i ” to a particular Destination facility “j ” is given as: DP [i]Gravity[j] =
Equation 7
C[j]weightα eβ·dist[i,j]
where dist[i,j] is the network distance from Origin “i” to a particular Destination facility “j ”. The sum of accessibilities from the demand point “i ” to all available Destination centers “j ” that are available within the specified Search Radius r around it, is given as: DP [i]Gravities =
#c
DP [i]Gravity[j]
j
Equation 8
The probability that a person at Origin point “i ” will patronize a particular facility at Destination point “j ” is given as a ratio between i’s Gravity to that particular Destination divided by i ’s sum of Gravities to all available Destinations, including “j ”:
DP [i]P robability[j] =
DP [i]Gravity[j] DP [i]Gravities
Since the Gravity calculations are determined by both the proximity of the Destinations and the attractiveness (weight) of the Destinations, the probability of visiting any destination is determined by both factors simultaneously as well as the presence of competing facilities (Equation 9).
URBAN NE T WORK ANALYSIS
111
Given these definitions, the tool can output two different results, depending on if the user turns on the ApplyImpedance option. If ApplyImpedance = On, then patronage at center “j ” is calculated as follows:
#DP
C[j]P atronage =
DP [i]W eight · DP [i]P robability[j] ·
i
1 eβ·[i,j]
This solution reflects the “gravity discounted patronage” at each Destination facility “j ”. If empirically calibrated alpha and beta values are used, then the result can be used to estimate the actual number of patrons visiting each facility. Due to the last “distance decay” term in Equation 10 (1/e β·dist[i,j ]), not all demand weights reach Destination facilities, resulting in lower total patronage across all Destination than the sum of Origin weights would suggest (Sevtsuk and Kalvo 2017). if ApplyImpedance = Off, then the last term in the equation is dropped and patronage at center “j ” is calculated as follows:
Equation 9
Equation 10
#DP
C[j]P atronage =
DP [i]W eight · DP [i]P robability[j]
i
This solution reflects the traditional Huff (1963) model, where a share of total demand weights is allocated to each center “j ” and the sum of estimated patronage results across all Destinations equals the sum of Origin weights that are located within the specified Search Radius from Destinations. 4 . 4 . 8 . DISTRIBUTE
WEIGHTS
The Distribute Weights tool re-distributes point weights from Origin locations to newly created point sequences along network routes that lead to Destinations. This can be useful for modeling en route demand for facilities, where demand from Origin locations is distributed as if they were walking along routes that lead to their Destinations. If Origin weights indicate the “number of households in residential buildings”, for instance, then the tool can re-distribute these weights from buildings to walking routes that lead nearby transit stations, dropping new points at chosen distance intervals (e.g. 50m). The total sum of weights on the newly created distributed points stays the same as the sum of weights at Origin locations.
Figure 83: Distribute Weights tool
112
analysis tools
Figure 84: The figure shows distributed weights from the two buildings in blue to the bus stop with a 10m interval. In this example, the number of residents in these buildings are designated as the demand weights. The weights are combined when the two routes converge. O rigins: building in blue D estinations: bus stops in red Detour ratio: 1 Weight: residents Coefficient: 1 0.94 0.70 0.24 Distributed weights
10m intervals
14 residents
4 residents
Consider an Origin point “O”, which has a weight of “5” people, who are assumed to walk to a Destination point “D”. If we used the Betweenness tool to estimate these walks, then each route segment along the shortest path would get a Betweenness result of “5”, indicating that it is passed by 5 people. This is useful for predicting the number of passersby at each network segment or Observer point along the way. But if we wanted to use these values to estimate demand for ice cream in a neighborhood, for instance, then the Betweenness values would not be suitable. Each Observer point would have the same Betweenness value — “5” in this case” — and if there are 12 Observers along the shortest path between “O” and “D”, Betweenness values would suggest that demand for ice cream is made up of 12 × 5 = 60 people, not 5 the we started out with. The Distribute Weights tool, instead, allows us to distribute the Origin weight of “5” equally to multiple Observer points along the way, such that the sum of the Observer values still equals “5”. Each of the 12 Observer points obtains a value of 5/12 = 0.416. If this sequence of distributed demand points is used to model demand for ice cream, then the total demand is still “5” people. The tool requires Origin points, Destinations points as well as Observer points, the latter of which receive the redistributed Origin weights. If you don’t have a pre-defined set of Observers available, the tool offers an option to create Observer points for you at a chosen distance interval, placed along your network segments. Note, that new Observer points are only created along segments that are used by trips on the network. Segments that are not crossed by any trips, do not receive Observer points. Observer points created by the tool are placed on the active Rhino layer.
URBAN NE T WORK ANALYSIS
113 Figure 85: This analysis shows distributed weights from all buildings to transit stations in Punggol, Singapore. There are three types of Destinations, each receiving a different proportion of the original demand weights. This example assumes 70% of households go to MRT, 20% to LRT, and 10% to bus stops. 90
90 90 90
Once Origins, Destinations and Observers are given, the tool offers the following options on the command line: 90
90
132
O rigins: all buildings
112
112
90
132
90
90
90
90 90
Distribute weight (Search=Nearest DetourRatio=1 Weight=Count SaveOptions Coefficien t=1 CleanTouchedWeights=True): Search = Nearest option defines which Destinations are used for each Origin point — only a single Nearest Destination, All Destinations, or all Destinations that are within a given Search Radius from the Origin. Detour Ratio = 1 option defines which travel routes are used to go from Origins to Destinations. The input describes the ratio between allowable path lengths to the shortest available path length between any Origin – Destination pair and it is constrained between “1” and “2”. In case of the default value “1”, allowable path lengths are the same as the shortest path and only shortest paths are used. If it is “1.1”, then all paths that are up to 1.1 times the shortest path length are used to travel to the Destination (up to 10% longer than shortest path) and so on. Weight = Count option defines which weight at the Origin points should be redistributed to Observer points. The default option “Count” will simply split a value of “1” from each Origin to all Observers between an Origin – Destination pair. Using Weight = Count will produce a new weight for Observer points called “Count_Fraction”. Save Options enable you to define a new name for the re-distributed weights, which are saved to the Observer points.
D estinations: all transit stations Detour ratio: 1 Weight: residents Coefficient: 1 bus stops LRT stations MRT stations 388.00 71.82 0 Distributed weights
114
analysis tools
To use a custom name, click on the option and set Use = True and type in a custom name (e.g. “d_weight”). Using a custom name allows you to keep track of multiple weight distributions. It also allows you to iteratively append multiple weight distributions to the same Observers and weight names if you set CleanTouchedWeights = False (see below). = Coefficient 1 variable determines what portion (%) of the Origin weight is redistributed to Observers. If you keep the coefficient at the default “1”, then 100% of the Origin weights are redistributed. If you set it to “0.5”, then only half the weights are redistributed to Observers, in which case the total sum of redistributed weights will equal half the sum of original weights you started with. Reducing the coefficient below “1” will not detract values from the original weights at the Origin points. CleanTouchedWeights = True option determines whether (True, False) weights at Observer points carrying the same weight name are over-written with new results or not. If set to True, then previous results are over-written. If set to False, then new results are appended to previous results, allowing you to cumulatively add weight distributions to Observers. This can be useful, when you intend to successively model different types of trips from the same Origins. For instance, if you model trips from households to MRT stations and bus stations, you could allocate 60% of trips to each household’s nearest MRT station (using Co efficient= 0.6) and 40% of trips to each household’s nearest bus stop (using Coefficient= 0.4), using the same SaveOption name (e.g. “transit_walks”) on both operations, and set CleanTouchedWeights = False. In this case, the same Observer points would cumulatively gather distributed weights from both MRT and bus walks, saved to the Observers with a weight name “transit_walks”. See the example case study from Singapore in Section 2.2 for an applied illustration of the use of this tool. 4 . 4 .9. CLUSTERS Figure 86: Clusters tool
Cluster Radius <800> (MinClusterSize=10 GroupAndCopy=On SaveWeight=Off): The Clusters tool finds spatial clusters of points along networks. It can be used to detect business clusters, street vendor clusters, crime clusters, wealthy household clusters, tree clusters, or any other event aggregations on networks. Clusters are defined according
URBAN NE T WORK ANALYSIS
115 Figure 87: This figure shows the three Clusters in the extent of the map that have a minimum cluster size of 10 where each member of the Cluster is at most 50m away from another member. Origins: all businesses Destinations: all businesses Cluster radius: 50m Min cluster size: 10 Figure 88: This figure shows the Clusters of the map that have a minimum cluster size of 25 where each member of the Cluster is at most 75m away from another member. O rigins: all businesses Destinations: all businesses Cluster radius: 75m Min cluster size: 25
116
analysis tools
to two user defined criteria: 1) a minimum number of points that constitute a cluster, and 2) the maximum allowable network distance from each member of the cluster to at least one other member in the same cluster. For instance, business clusters can be defined as an aggregation that contains at least 25 businesses, where the maximum distance from each member is no more than 75 meters to at least one other member in the same cluster. In order to run the tool, a network is required (using Add Curves to Network), and the same set of points have to be added to a network as both Origins and Destinations (using Add Origins and Add Destinations tools). The options on the command line include the following: Cluster Radius defines the maximum allowable network distance from each member of the cluster to at least one other member in the same Cluster. You can type the desired distance to the command line. Note that the units of distance used follow the Rhino drawing units in your file. MinClusterSize = 10 defines the minimum number of points that constitute a Cluster. GroupAndCopy = On triggers whether the points that satisfy the above two Cluster criteria are grouped by cluster and copied to the active layer. SaveWeight = Off option determines whether cluster numbers are recorded as Cluster weights to the original input point attributes. This can be useful, for instance, for copying the weights out to Excel, counting how many members each cluster has or analyzing the Clusters composition in other table manipulation software. Note that only points that belong to a Cluster obtain a Cluster attribute weight.
118
analysis tools
URBAN NE T WORK ANALYSIS
4.5.
GRAPHICS
119 Figure 89: Mineral color pigment collection at the Harvard art museum
The last section of the UNA toolbar offers users an ability to control how different analysis results or weights are displayed on the screen, or to prepare UNA result graphics for further manipulation in other software, such as the Adobe Creative Suite. 4 . 5 .1
GRAPHIC OPTIONS
This tool allows you to change the graphic appearance of UNA features: color schemes of results, result labels, point connection lines etc. The following options are available on the command line: Graphics (Color=WhiteRed Results=RedundancyIndex Weight=None Mode=Result Node=On NodeId=Off NodeD2=Off DotConnections=On DotArrow=Off DotLabel=On DotId=Off Dots=On Edges=On EdgeLabels=On Font=12 DotSize=5): Color = WhiteRed allows you to choose different coloring schemes for visualizing UNA analysis results or Origin / Destination / Observer attribute weights. Results = RedundancyIndex option controls which analysis result is currently being displayed. Weight=None option controls which point weights (numeric attributes only) are currently being displayed. Note that point weights can only visualized when a) you have added the points to a network as Origins, Destinations or Observers and b) when Mode is set to Weight, as discussed in the next option. Mode = Result option toggles whether the graphics display analysis results or numeric object weights. Node = On option determines whether small black cross symbols are drawn at the dead-end nodes of the network. These can be useful for detecting topology problems in a network. If a black cross appears at a T-intersection where you expect all paths to connect, then you know there is a problem. Most likely, the street entering perpendicularly does not share an endpoint with the other two segments. Fix the topology errors by manually editing network segment endpoints.
Figure 90: Graphic Options tool
120
graphics
NodeId = Off toggles Rhino Node IDs On or Off from the display mode. NodeIds are network nodes with automatically assigned values that the user typically does not need to see. NodeD2 = Off option turns On/Off red warnings with a numeric value “2” that tell you which nodes have a degree centrality “2” — these are nodes where only two network segments meet. If you see red number “2” at an intersection you expect more than two links to connect, you know you have a topology issue. See Add Curves to Network tool in Section 4.3.1 for more information. DotConnections = On option controls whether blue, red and gray connection lines from Origin, Destination and Observer points respectively are shown to the closest network edge. DotConnections effectively visualize the locations, where Origins, Observers or Destinations are assumed to connect to the network. DotArrow = Off option allows you to further add blue/red/ gray arrows to the network representation to show the direction of trips at each node. DotLabel = On controls whether numeric UNA analysis results are shown next to nodes or not. This is an important feature — turning DotLabels off will hide the analysis result labels from display. If you operate with a large number of analysis points, then turning the DotLabels = Off helps improve the display response rate and navigation in the Rhino model. DotId = Off controls whether UNA ID numbers for each network node are displayed. Typical users do not need to display these numbers. Dots = On controls whether network nodes with colored results are shown or hidden. Edges = On controls whether color-coded results are also shown on network edges. This control only affects the Betweenness analysis, where results can be visualized at the edge level. EdgeLabels = On option allows you turn numeric results On or Off from network edges. This control only affects the Betweenness analysis, where results can be visualized at the edge level. Font = 12 controls the size of the font on result labels on the screen. DotSize = 5 controls the size of the dots where UNA results are shown.
URBAN NE T WORK ANALYSIS
4 . 5 . 2. BAKE
121
WEIGHT COLOR
The Bake Weight Color tool allows you to assign UNA result colors that are visualized on the screen to the source objects as Rhino object colors. This can be useful if, for instance, you wish to export UNA result graphics out to other graphic design software, such as Adobe Illustrator, AutoCAD, etc. The tool only works with point objects. Betweenness analysis results, shown at the edge level, cannot be baked to the underlying Rhino curve objects. To Export Betweenness results, use Observer points. If you wish to ultimately present Betweenness results on edges in Adobe Illustrator drawings or other graphic design software, a common work-around is to first Export Betweenness results at the Observer points out to ArcGIS and then join the results to network edges in ArcGIS using a Spatial Join, giving each edge the average of its Observer values. Because Export/Import procedures between Rhino and ArcGIS can result in small deviations in point locations, it is good practice to first give a small buffer (e.g â&#x20AC;&#x153;2 metersâ&#x20AC;?, depending on the type of network) to the Observer points in ArcGIS, before performing the spatial join operation. See the Frequently Asked Questions section for more information on curve exports to GIS.
Figure 91: Bake Weight Color tool
URBAN NE T WORK ANALYSIS
5. FREQUENTLY ASKED QUESTIONS
123
124
graphics
INSTALLATION Where can I download the UNA toolbox? The UNA toolbox can be downloaded from multiple repositories: City Form Lab website: http://cityform.gsd.harvard.edu/projects/una-rhino-toolbox Bitbucket: https://bitbucket.org/cityformlab/urban-network-analysis-toolbox/downloads/ Food for Rhino: http://www.food4rhino.com/app/urban-network-analysis-toolbox UNAToolbox.rhi file does not install, when I double-click on it? With some recent McNeel updates the .rhi installer for UNA may not be recognized as a program on windows if you double click the “UNAToolbox.rhi” file and navigate through the installation steps. This is a known Rhino bug that McNeel is aware of. You can still install the UNA toolbox in one of two ways: a) drag and drop the .rhi installer file into an open Rhino window (this is the easiest solution) or b) re-associate the *.rhi file in Windows with C:\Program Files\Rhinoceros 5 (64-bit)\System\x64\rhiexec.exe. I have installed the UNAToolbox.rhi file, but the UNA toolbar doesn’t appear in my Rhino window? If you have successfully installed the “UNAToolbox.rhi” file and don’t see the toolbar in a Rhino window, restart Rhino, go to Rhino, Tools > Toolbars tab and make sure to check the box next for the UNA Toolbar. Then Click “OK” to close the Toolbar window. I have installed the UNAToolbox.rhi file, restarted Rhino, but the checkbox for UNA toolbar doesn’t even appear in the Rhino Toolbars window? Make sure you are installing the UNA toolbox on an official version of Rhino 5 or newer and have the latest updates installed. Updates can be installed from Rhino help > check for updates. The UNA toolbox does not work on pirated copies of Rhino.
URBAN NE T WORK ANALYSIS
CREATING NETWORKS I have built a network from curves and added Origin and Destination points, but my network returns unexpected results? Make sure your network is made of individual edge segments that share endpoints at nodes, where two or more curves meet. UNA analyses can only route trips from one segment to another if they share an endpoint. However, intersecting curves without a common endpoint can be used to model overpasses and underpasses, where trips cannot connect from one segment to another. See section 4.3.1. for more details about creating networks from curve objects. I added points to a network, but some of my points are marked with red crosses? If some of the points you added to a network are marked with red crosses, it means that they are too far from nearest network edges to be added. If you think they seem close enough and should be added, you can try to move the points closer to the network or add network segments closer to the points. Occasionally, this might also signal that you have problems with the network itself. This can mean that the network is broken down to very little segments, not only at actual path intersections, but also at individual points along what should normally be a continuous curve. See the next question on how to solve the latter issue. When I add points to a network, the points do not connect to their nearest segments, but instead to segments that are further away? For the UNA toolbox to effectively find the nearest network segment for each Origin, Destination or Observer point, the algorithm handles all segments as rectangles, whose diameter is several times larger than the length of the segment. This allows it to find all candidate lines using the R-Tree algorithm, from which distances are evaluated to determine the nearest points. The scenario, where this procedure can lead to an incorrect nearest segment allocation arises when:
125
126
graphics
1. 2. 3.
There are a lot of short segments, whose search area is small. There is a longer segment further away, whose search area is large and to which points are connected instead. A point is more than a few diagonals away from all small segments.
There are several solutions to this issue: A. Join the small network segments together into longer polylines, only having line breaks at intersections. This is generally a good practice to follow anyway, so that network drawings are clean and only show nodes at intersections, where segments cross and connect. B. You can also use the function unaBindEdge from the command line to determine which edge a particular point should be assigned to. You need to do this before you add points to the network. C. You can move points closer to the network, so that they are within a couple of diagonal lengths from the network segments. Since the toolbox functions independent of drawing units (people can bring in drawings in meters, feet, kilometers etc.), we canâ&#x20AC;&#x2122;t tell the algorithm to look for points in a given radius (e.g. 25 meters), but have to instead rely on segment proportions. How can I make sure my network is topologically connected and correctly prepared for UNA analysis, without having to go over every node by hand? Networks can be composed of any curve elements in Rhino (e.g. lines, polylines, arcs, splines) and form both two-dimensional and three-dimensional lattices. In preparing networks, it is important to split or explode curves, where they meet other curves â&#x20AC;&#x201D; network routing between two line-segments only works if the segments share a common endpoint. If you downloaded the data from US Census Tiger shapefiles and exported them from ArcGIS to DWG files, the network might already be well built and further editing might not be needed. Always inspect the network visually, before running the analysis. Turning on Nodes in UNA Graphic Options allows you to visualize, where you have disconnected or dead-end line segments. If your network requires fixing, the following Rhino curve editing procedures can allow networks to be rapidly prepared for UNA analyses:
URBAN NE T WORK ANALYSIS
1. 2.
Use the Join tool to join all curve segments with each other. Use the Intersect tool to generate points wherever two or more curves intersect. 3. Use the Split tool to split all curves with the above intersection points. This should quickly produce a planar network where curves are continuous, and only broken up where they intersect with other curves.
127
128
graphics
RUNNING UNA ANALYSES I have added Origin and Destination points, but when I try to run a UNA analysis tool, nothing happens? Make sure you have built a network using the Add Curves to network too. See Section 4.3.1. for more information on creating networks. When I try to run a UNA analysis tool, it seems to crash midway and doesn’t finish the analysis? Make sure you have no duplicate objects (Origin, Destination or Observer points or network segments in the Rhino scene. Rhino has a handy tool called Select Duplicate objects, which picks out all overlapping geometries. Delete overlapping objects, where appropriate. When I run UNA analyses, some of the buildings that I expect to obtain results or get selected, are not selected or do not obtain results? This typically happens if your network is not properly connected. For instance, a network segment where the Origins or Destinations that do not get expected results are located, might be disconnected from the surrounding network. You may want to turn on the “Nodes” in Graphic Settings (see section 4.5.1.), which will mark all the locations where you have “dead ends” or nodes with “degree two”, in the network. If you observe dead ends at intersections or on segments that shouldn’t have them, zoom in close and visualize the actual curve end points by turning “Points On” in Rhino. Fix the network topology issues and try again. I am trying to run Betweenness analysis, but it is taking very long and the progress seems slow? Betweenness analysis is rather computation intensive and the speed of its execution depends directly on the number of Origins and Destinations you input, the connectivity of the network, as well as Search Radius and Detour Ratio parameters you specify. First, there are a couple of analysis parameters that you may be able to
URBAN NE T WORK ANALYSIS
adjust to streamline the analysis. If you are using a relatively high Detour Ratio (e.g. 1.2 – 2.0), try reducing the Detour Ratio to “1.1” or “1”. The analysis is fastest with a Detour Ratio of “1” – when no alternative routes between Origin-Destination pairs have to be found. Second, depending on the nature of your analysis, you might also consider reducing the Search Radius. This will limit how far Origin-Destination pairs can be from each other to be still included in the anlysis. Note that the Search Radius is measured in the same units as your Rhino drawing. If these settings do not solve your issues, the problem can be also reduced by decreasing the number of Origin and Destination points you use. For instance, having “5” Destinations and “1,000” Origins yields “5000” route calculations for the Betweenness tool, which should typically not pose a problem (unless you require significant Detour Ratios to be applied). But having both “1,000” Destinations and “1,000” Origins already yields “1,000,000” route calculations, which is 200 times more. This increases exponentially if you add Detour Ratios to the calculation. On a decent desktop machine, having “1,000” Destinations and “1,000” Origins should still be able to run. But the more Origin-Destination pairs you include, the slower the analysis becomes. We have been able to run up to around 10,000 Origins and 10,000 Destinations on a relatively powerful desktop computer. If your Origin Destination data contains thousands of individual house locations, you might consider instead running the analysis between street intersections – using intersections as both Origins and Destinations. This is demonstrated in the Cambridge and Somerville case-study in section 2.3. Generating end points for all network segments is also further explained below. How can I generate end points for all network segments, so I can run the Betweenness analysis between all street intersections? First, join all network curves to each other, using the Join tool in Rhino. Next, generate all curve start points from your network, by using the crvStart command on the Rhino command line and selecting your network. Repeat this for all curve end points, by using the crvEnd command on the Rhino command line and selecting your network. These latter steps are necessary to guarantee that you get points at the ends of your network, not only at the intersections. Then, generate all curve intersection points, by using the Intersect tool (or type Intersect on the command line)
129
130
URBAN NET WORK ANALYSIS TO OLBOX
and select your network. Note that many of the curve start or end points may now overlap with intersection points. Use the select duplicates tool (or type SelDup on the command line) to select points that are duplicates of other points. Delete the duplicates. This should produce a set of points at all curve start, end and intersection points that you can use as Origins, Destinations or Observers in UNA analyses. How can I take the Betweenness analysis results of individual path segments (not points) out to GIS? The UNA Export function only works for points. In order to get the line segment results to GIS, it is usually handy to 1) run the Betweenness analysis with Observer points, with enough points so that every network edge has a couple, 2) export the Observer results out to a table with X and Y coordinates, 3) paste the results to excel and save them as a “CSV” table, 4) add the “CSV” table to ArcGIS, 5) right-click on the “CSV” layer in GIS and use the “Add X,Y coordinate data as layer” tool. This will recreate the Observer points in ArcGIS, keeping the Betwenness results associated with them. Note that your projection system in GIS should match the units that you used in Rhino. Finally, you can perform a spatial join between the GIS Observer points and network lines. Since the above import procedure can produce small displacements for points, it is usually good practice to first add a reasonable buffer around the X,Y points you recreated in GIS, and then perform a spatial join between buffered points and network lines, associating every network segment with an “average” value of Observer points it intersects with. Once you have the Betweenness results for line segments in GIS, you can use any of the GIS symbology tools to visualize the line values. How can I save the color coding from UNA analysis results out to point objects, so that the colors remain, even when I delete my network? Use the “Bake Weight Color” tool at the right end of the UNA toolbar. You can then export the colored points out as a DWG file and import them to Adobe Illustrator or other graphic software packages.
URBAN NE T WORK ANALYSIS
How can I export my color-coded analysis points out to Adobe Illustrator? Only using the DWG file format works for transferring points to Adobe Illustrator. The .ai format does not, because it does not recognize point objects. Export your points out from Rhino to DWG. When you open the DWG file in Illustrator, point objects may not immediately appear because they are infinitely small. Use the select all command in Illustrator and see if you see objects being selected. If they are, assign them a larger outline thickness to see the points. How can I take my Rhino UNA analysis results to Excel or GIS? See section 4.2. of the user guide. Where can I obtain street network data for my city? If you work in the U.S., the Census Bureau’s TIGER/Line Shapefiles website is a great resource: https://www.census.gov/cgi-bin/ geo/shapefiles/index.php To obtain street centerlines in any US city, choose “Roads” from the dropdown many, the year and state you wish and press Download. In the rest of the World, some countries also have official public downloads of street network available (e.g. the Ordnance Survey in the UK). At a global scale, OpenStreetmap (https://www.openstreetmap.org) is a terrific resource. Several third-party tools have been built to facilitate downloading streets from OpenStreetmap for your city or region. See for instance: http://geoffboeing.com/2016/11/osmnx-python-street-networks/ https://www.geofabrik.de/data/shapefiles.html You can also simply draw your own networks in 2D and 3D in Rhino based on satellite images, site plans, building plans etc.
131
â&#x20AC;&#x2030;
URBAN NE T WORK ANALYSIS
6. BIBLIOGRAPHY
133
134
graphics Bhat, C., Handy, S., Kockelman, K., Mahmassani, H., Chen, Q., & Weston, L. (2000). Development of an Urban Accessibility Index: Literature Review. (B. of E. R. Center for Transportation Research The University of Texas at Austin, Ed.). Austin, TX: The University of Texas at Austin. Boarnet, M. G., Joh, K., Siembab, W., Fulton, W., & Nguyen, M. T. (2011). Retrofitting the suburbs to increase walking: Evidence from a land use-travel study. Urban Studies, 48(1), 129–159. Cervero, R. and Duncan, M. (2003) ‘Walking, bicycling, and urban landscapes: evidence from the San Francisco Bay area’, American Journal of Public Health 93, 1478–83. City Form Lab, Hansen Partnership, & The World Bank. (2015). Surabaya Urban Corridor Development Program. Jakarta, Indonesia. Retrieved from https://www.dropbox.com/s/nefj1u5z1n26xzp/140714_Surabaya_Urban_Corridor_ Development_Program_FINAL.pdf?dl=0 DiPasquale, D., & Wheaton, W. C. (1996). Urban economics and real estate markets. Englewood Cliffs, NJ: Prentice Hall. Ewing, R. and Cervero, R. (2010) ‘Travel and the built environment’, Journal of the American Planning Association 76, 265–94. Freeman, L. C. (1977). A set of measures of centrality based on betweenness. Sociometry, 40, 35–41. Eppli, M., & Shilling, J. (1996). How Critical is a Good Location to a Regional Shopping Center? Journal of Real Estate Research, Vol. 12(3), 459–469. Garrison, W. L., & Marble, D. F. (1962). The Structure of Transportation Networks. (U. S. A. T. Command, Ed.) (pp. 73–78). U.S. Army Transportation Command Technical Report. Gehl, J. (1987). Life between buildings : using public space. New York: Van Nostrand Reinhold. Hansen, W. G. (1959). How Accessibility Shapes Land Use. Journal of the American Planning Association, 25(2), 73–76. Hensher, D. A. (2004). Handbook of transport geography and spatial systems (p. xxii, 672 p.). Amsterdam ; Boston: Elsevier. Hess, P. M., Moudon, A. V., Snyder, M. C. and Stanilov, K. (1999) ‘Site design and pedestrian travel’, Transportation Research Record 1674, 9–19. Hillier, B. (1996). Space is the machine : a configurational theory of architecture (p. xii, 463 p., [8] p. of plates). Cambridge ; New York, NY, USA: Cambridge University Press. Retrieved from http://www.loc.gov/catdir/toc/cam023/95021500.html Hillier, B., & Hanson, J. (1984). The Social Logic of Space. Cambridge: Cambridge University Press. Hillier, B., Hanson, J., & Peponis, J. (1987). Syntactic Analysis of Settlements. Architecture and Behaviour, 3(3), 217–231. Huff, D. (1963). A Probabilistic Analysis of Shopping Center Trade Areas. Land Economics, Vol. 39(No. 1), 81–90. Jaber, A. A., & Papaioannou, D. (2017). Benchmarking Accessibility to Services Across Cities. (Working Paper No. 2017–xx). OECD, International Transport Forum. Paris, France. Jiang, B., Claramunt, C., & Batty, M. (1999). Geometric accessibility and geographic information: extending desktop GIS to space syntax. Computers, Environment and Urban Systems, 23(2), 127–146. doi:10.1016/S0198-9715(99)00017-4 Kansky, K. J. (1963). Structure of transportation networks: relationships between network geometry and regional characteristics (p. x, 155 p.). Chicago, IL. MTI Economic Review Committee Sub-committee on Domestic Enterprises. (2002). Neighborhood Working Group Report. Singapore.
URBAN NE T WORK ANALYSIS Mohsenin, M., & Sevtsuk, A. (2013). The impact of street properties on cognitive maps. Journal of Architecture and Urbanism, Volume 37(Issue 4), 301–309. Okabe, A., & Shiode, S. (2001). SANET: A toolbox for spatial analysis on a network. Journal of Geographical Analysis, Vol.38(No. 1), pp.57–66. Okabe, A., & Sugihara, K. (2012). Spatial Analysis Along Networks: Statistical and Computational Methods (Statistics in Practice) (p. 296). Wiley. Retrieved from http://www.amazon.com/Spatial-Analysis-Along-Networks-Computational/ dp/0470770813 Porta, S., Crucitti, P., & Latora, V. (2005). The network analysis of urban streets: a primal approach. Environment and Planning B, 35(5), 705–725. Porta, S., Strano, E., Iacoviello, V., Messora, R., Latora, V., Cardillo, A., Scellato, S. (2009). Street centrality and densities of retail and services in Bologna, Italy. Environment and Planning B: Planning and Design, 36, 450–465. Pushkarev, B., & Zupan, J. (1975). Urban Space for Pedestrians. Cambridge, MA: MIT Press. Sevtsuk, A. (2010). Path and Place: A Study of Urban Geometry and Retail Activity in Cambridge and Somerville, MA. MIT, Cambridge. Sevtsuk, A. (2012). Analysis and Planning of Urban Networks. In K. A. Zweig (Ed.), Encyclopedia on Social Network Analysis and Mining. Springer. Sevtsuk, A. (2013). Networks of the built environment. In D. Ofenhuber & C. Ratti (Eds.), [de]coding the city - how “big data” can change urbanism. Birkhäuser. Sevtsuk, A., & Mekonnen, M. (2012). Urban Network Analysis Toolbox. International Journal of Geomatics and Spatial Analysis, 22(2), pp. 287–305. Sevtsuk, A., Mekonnen, M., Kalvo, R., & Amindarbari, R. (2014). Redundant Paths for Urban Network Analysis. Conference paper. ESRI Geodesign Summit. Sevtsuk, A. (2014). Location and Agglomeration: the Distribution of Retail and Food Businesses in Dense Urban Environments. Journal of Planning Education and Research, 34(4), 374–393. Sevtsuk, A., & Kalvo, R. (2017). Patronage of urban commercial clusters: a networkbased extension of the Huff model for balancing location and size. Environment and Planning B: Urban Analytics and City Science, 0(0), 1–21. Sevtsuk, A. (2018). Estimating pedestrian flows on street networks: revisiting the betweenness index. Paper presented at the American Association of Geographers Annual meeting in New Orleans, April 2018. Forthcoming. Speck, J. (2013). Walkable City. How downtown can save America one step at a time. North Point Press. Stibbs, R., & Tabor, P. (1970). The Evaluation of Circulation in Buildings: A Mathematical Model. Cambridge: Cambridge University. Tabor, P. (1976). Networks Distances and Routes. In L. March (Ed.), (pp. 366–367). Cambridge: MIT Press. Targa, F. and Clifton, K. (2005) ‘The built environment and trip generation for non- motorized travel’, Journal of Transportation and Statistics 8 (3), 55–70. Vragovic, I., Louis, E., & Diaz-Guilera, A. (2005). Efficiency of information transfer in regular and complex networks. Physics Review E., 71(026122). Xie, F., & Levinson, D. (2007). Measuring the structure of road networks. Geographical Analysis, July 2007
135
© City Form Lab, 2018 License The Urban Network Analysis Toolbox for Rhinoceros 3D is distributed by the City Form Lab and can be used under the License of Creative Commons Attribution-NoDerivatives 4.0 International. For more information, see: http://creativecommons.org/licenses/by/4.0/legalcode ISBN 978-0-692-17277-3 Contributors Andres Sevtsuk, Director, City Form Lab Assistant Professor of Planning, Harvard GSD. Raul Kalvo, researcher City Form Lab UNA Rhino Software development. Kevin Chong, researcher City Form Lab Analysis and Graphics. Hao Ding, researcher City Form Lab Case study analysis and writing. Stuudio Stuudio Graphic design and layouts. Acknowledgements We are thankful to the Harvard GSD Dean’s office for supporting this publication.
City Form Lab