Content uploaded by Kimon Krenz
Author content
All content in this area was uploaded by Kimon Krenz on Jan 04, 2018
Content may be subject to copyright.
Proceedings of the 11th Space Syntax Symposium
150.1
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
KIMON KRENZ
Space Syntax Laboratory, The Bartlett School of Architecture, UCL
kimon-vincent.krenz.12@ucl.ac.uk
#150
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN
SPACE SYNTAX ANALYSIS
ABSTRACT
The application of volunteered geographic information has rapidly increased over the past
years. OpenStreetMap (OSM) forms in this context one of the most ambitious and promising
projects, providing consistent global coverage of street network information. With a constantly
growing number of participants and the implementation of governmental and proprietary
based information is a complete coverage of global street networks within reach. The data
allows comparative cross-country analyses and any method developed within its framework
are transferable to other cases. This makes OSM a powerful and desirable data source for
applied network analyses, such as space syntax. However, OSM data does not come without
obstacles. Inconsistent representation of space, topological fragmentation and accuracy are
just some of the problems that one faces when employing OSM data. In fact, without prior
processing and simplication of the network, results dier signicantly between case studies.
This paper presents a method for OSM data set simplication as well as the theoretical and
analytical reasoning behind it. The simplication is done by a series of ArcGIS workows and
algorithms. The outcome of this process is compared to an angular segment analysis (ASA)
of a segment model, an Integrated Transport Network (ITN) Ordnance Survey data model
and an OSM street network data model. The results show that a simplied version of OSM
data is highly comparable to a segmented axial line representation and that such data sets
constitute an appropriate alternative for situations where segment maps are not available, such
as complex, large-scale regional models and cross-country comparisons. The simplication
workow is transferable to other cases and data sets and helps overcoming common problems
while signicantly improving computational time needed in the process.
KEYWORDS
Volunteered Geographic Information, Open Street Map, ArcGIS, Space Syntax, Street Network
1. INTRODUCTION
The aim of this paper is to present a workow and methodology that allows the use of
OpenStreetMap (OSM) data in space syntax angular segment analysis (ASA). The reasoning
behind employing such data sets is the increasing scale of analytical investigations in the
context of space syntax. This augmentation of scale has become particularly necessary due
to the extensive global growth of cities and their urban hinterland into large complex urban
regions. These urban structures are simply too vast to be mapped manually or generated by
automated algorithms. This has created a situation in which the time and economic feasibility
of traditional as well as algorithmically derived axial line maps needs to be revisited. Previous
research proposed to make use of governmental so-called road-centre line data as an alternative
for a segmented axial line, more commonly referred to as segment maps (SM). However, very
little has been said about the disadvantages of such approaches particularly when global
Proceedings of the 11th Space Syntax Symposium
150.2
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
comparability is needed, something in which space syntax is believed to be particular strong.
OSM road-centre line data, on the other hand, I will argue, forms not only an appropriate
alternative basis for models in these situations, but it also allows global comparability as well
as being freely accessible on a large scale. Nevertheless, OSM data does not come without
disadvantages either. Particularly concerning excessive information in such data sets, which
makes a simplication prior to any ASA application necessary, caution needs to be exercised.
This paper consists of three parts; the rst revisits the foundation of space syntax axial line
models and the sequentially developed analytical method of ASA and its segment map (SM)
model. An emphasis is placed on the model underlying the analysis and the diculties arising
in the model generation generally and in large-scale applications particularly. In this light,
volunteered geographic information and governmental road-centre line data, such as the
British Integrated Transport Network (ITN) are reviewed as alternatives for SM models. Finally,
advantages as well as disadvantages of OSM data are discussed and the eect of these on ASA
outcomes.
The second part presents the structure and particularities of the previously introduced OSM
data, as well as the diculties researchers are facing when employing such data in ASA. I
discuss the three main diculties, which are topological inconsistency, trac management
components and excessive or redundant nodal information. I propose dierent GIS strategies
to simplify and remove such redundant information and explain the theoretical reasoning
behind them. The result is a newly derived simplied OSM network model, termed ‘SIMP’.
The third part evaluates the new SIMP model against OSM, ITN and SM models in ASA. I do this,
using descriptive statistics, visual comparisons, as well as a Pearson and Spearman correlation
analysis. The results show an overall high correlation between the four models, conrming
previous ndings. The new SIMP model exhibits higher correlations with the SAL model than
both OSM and ITN network models, indicating that a simplied OSM network does not only
form an appropriate alternative but one that presumably incorporates fundamental network
characteristics of SM models.
2. AXIAL MODELS AND ANGULAR SEGMENT ANALYSIS
Axial analysis forms one of the fundamental techniques of space syntax. At the core of an axial
analysis methodology lies the axial line map, a representation of the continuous structure of
open spaces in urban settings. The rst axial line model was introduced by Hillier and Hanson
(1984, p. 17) during the early 1980’s and dened as a system of fewest and longest intersecting
lines covering all open spaces. These lines are the result of a two-step process where the spatial
system under investigation is rst represented through a two-dimensional organisation of
convex spaces. Convex spaces are polygonal representations of continuous open spaces, in
which each part of a space must be visible from every other part. The underlying rule for drawing
a convex space is that each polygon must feature the best ‘area-perimeter ratio’, starting with
the ‘fattest’. In a subsequent second step, this system of convex spaces is covered by a one-
dimensional set of axial lines. Axial lines are linear representations of longest lines of sight and/
or movement. Each convex space must be covered by at least one axial line, while each line
needs to be the ‘longest straight’ line possible (ibid., p.17).
Although Hiller and Hanson describe this process as reproducible and objective, there is
some discussion and ambiguity about the comparability and making of axial maps. Problems
arise for instance with dierences in the level of detail or resolution in which convex spaces
are produced, as this impacts the number and distribution of the resulting axial line map.
Problems also arise with the diculty to arrive at comparable reproducible solution for the
same given urban context. Peponis et al. acknowledge in this regard ‘SpaceBox’1, a software
that automated the generative process of convex spaces, but they criticise the mathematical
1 SpaceBox is a software developed by Sheep Dalton (1988) and includes several space syntax related functionalities
one of which being the generation of an all convex space map. The software’s partitioning algorithm extends a
walls surface area collinear until the produced line reaches another wall surface. See Carranza and Koch for more
recent work on convex spaces (2013).
Proceedings of the 11th Space Syntax Symposium
150.3
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
rigour of its computational algorithms to generate convex spaces (1997, 1998). According
to Peponis et al. neither the initial principle of generating convex spaces based only on an
economic partitioning, nor the extension of surfaces to the next opposite wall is a sucient
method. Both lead to multiple, conicting solutions, implying that a more sophisticated set of
rules is necessary. Interestingly, although the methodology of convex spaces is thought of in an
urban context most of the discussions are set in the context of buildings. This might be due to
the time-consuming process of producing convex spaces for entire cities, with the sole purpose
of deriving an axial line map. The scale of the area under investigation and respectively the
time necessary to produce such convex representations is certainly one of the most important
inuencing factors.
Moreover, Desyllas and Elspeth argue that not only the production of convex spaces, in general,
is dicult, but that it constitutes a ‘mathematically impossible problem’ to link all maximal
convex spaces with axial lines in an identically repeatable manner (2001, p. 27.6). The core
problem here is that there are several solutions to axial lines that full the criteria of being the
longest as well as covering all convex spaces (Batty and Rana, 2004; Ratti, 2004). As a solution
to this technical and theoretical problem Turner et al. (2005) – building on an initial but not
ideal solution from Peponis et al. (1998) – proposed an automated methodology that produces
a fewest line axial map. The starting point of their method is vector information of open space
boundary polygons. Based on this, a so-called ‘all-line map’ is generated (Penn et al., 1997).
The ‘all-line map’ is a map that features all lines that connect each vertex of boundaries and
buildings with all other visible vertices, i.e. all possible lines of movement. In a following step
Turner et al. employ an algorithm to reduce this ‘all-line map’ to a fewest line axial map. Their
results are reproducible and strikingly similar to the original Hillier and Hanson axial map (2005).
However, his method of the fewest line axial map generation, does not constitute an appropriate
way to produce models for cities and regions. There are two primary factors, which prevent the
application in a citywide and regional context. The rst starts with the source of data and its
denition of open space, a problem that the very initial convex space methodology already
inherited. What to include and what to leave out in a graphical representation of the real world
is left to the individual cartographer or researcher and forms core challenges in comparative
cartography and map-making in general. This challenge is of particular importance when
investigating suburban or rural areas. Suburban and rural areas often lack a continuous urban
form and hence a given limitation for movement and visibility. Consequently, the denition of
what can be considered an ‘accessible open space’ becomes vague. A problem that researchers
are also facing in the context of developing countries exists as roads are often not solidied
and boundaries between public and private spaces are less established. In these cases, an
alternative could be to rely on other sources of geographic data of open spaces that follow
precise denitions. Such sources are for example governmental agencies for cartography,
geodesy and planning or volunteered geographic information, both of which have precise
denitions of what and how open spaces are mapped.
Computational time constitutes the second diculty. With a rising number of mapped open
space polygons and their vertices, the necessary computational time to generate the fewest
line axial map increases as well. Turner et al. give an account of the computational time needed
for their algorithm to compute fewest line axial maps. A model of the small town of Gassin
took 119 seconds to compute and featured 5217 lines in its initially generated all-line map and
38 axial lines in the nal result (ibid.). Thus, the computational process for an entire city or
even a region, with far more than one million street segments will take signicantly longer2.
While theoretically the algorithm could run for any time needed, in praxis this is limited by the
software design dealing with large data sets. Currently the most commonly used software for
this is depthmapX. Initial tests using the software on large urban systems generating fewest line
axial maps have consistently produced application crashes. Varoudis et al. state the maximum
number of segments that can be computed by depthmapX as <1.500.000 (2013), resulting in an
axial line map of approximately 15000 lines. This makes an automated generation of axial lines
for a metropolitan or regional system at the time not possible.
2 The total number of axial lines in cities with a population of 300,000 can range between 10,000 and 15,000.
Proceedings of the 11th Space Syntax Symposium
150.4
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
2.1 ROAD-CENTRE LINES AS ALTERNATIVE FOR SEGMENT MAPS
Initially, the focus of axial line maps was to have a tool that allowed understanding complex urban
systems in a simplied comparable manner. Over time the primary use of this morphological
descriptive tool was to be found in investigations into the deep relation between human
behaviour and space. From the development of the methodology, throughout the last 30 years,
researcher have consistently found correspondence between the topological relationships of
spatial systems and pedestrian movement (Hillier et al., 1993; Penn et al., 1998; Desyllas and
Elspeth, 2001; Hillier and Iida, 2005) as well as vehicular movement activities (ibid.; Turner,
2005; Law and Versluis, 2015; Serra, Hillier and Karimi, 2015) and even global transportation
networks (Hanna, Serras and Varoudis, 2013). This is particularly the case since the introduction
of ASA in space syntax as an extension of axial analysis (Turner, 2001). The emphasis thus shifted
from a theory and tool to analyse spatial congurations to one of predicting the potential of
human behaviour in the form of movement and ows. Four studies focus on alternatives that
constitute possible models for an analysis of movement and ows in the build environment:
The pioneering work by Thomson (2003), Dalton et al. (2003), Turner (2005, 2007) and following
up on these studies most recently the work by Dhanani et al. (2012). All authors investigate the
possible application of dierent types of so called road-centre line data. The reasoning is that
their approach relies on replacing a segment map, which is used in angular segment analysis
rather than the in traditional axial line model the SM is based on. This study will follow the path
taken by the above named researchers and base the comparison on a segmented axial line
model, rather than emulating an axial line model, which inevitably will later be segmented in
order to perform ASA.
Road-centre lines ideally represent the geographic centre of the public rights of way network,
a transportation network of all paths on which the public have a legally protected right to pass
and re-pass. These transportation networks are based on vector line information and can be
generated through a variety of GIS methods such as automated processes of on ground collected
GPS data, generative processes based on cadaster boundary data or manual tracing of roads on
aerial photographs. In a subsequent step, additional information can then be attributed to this
line information such as road names, road type, travel direction, road geometry information as
well as a large variety of other possible attributes.
This makes road-centre line maps a powerful tool for a variety of GIS based applications. The
ones applied the most are transportation modelling and navigation routing. Road-centre line
data was rst provided by local governments, such as the TIGER3 data set by the United States
Census Bureau or the ITN4 by the British Ordnance Survey, as well as commercial companies,
such as the Dutch Company TeleAtlas5 or American-based Company Navteq.6 The latter provides
mainly line-based data for navigational systems. With the rise of the Internet and Web2.07,
publicly accessible road centre-line information became largely available through dierent
sources. The most predominant sources are Google maps and Bing maps, both available under
restricted license for non-commercial usage. In contrast to governmental and proprietary
based information with restricted license stands volunteered geographic information (VGI).
VGI describes all geographic data, which is created, assembled and disseminated voluntarily
by individuals (Goodchild, 2007). Open source VGI projects such as OpenStreetMap (OSM) and
MapQuest are available under a GUP license and hence freely accessible to anybody. Due to the
increasing number of online participants all over the world these projects are on the rise and
establish a commercially as well as academically meaningful alternative.
3 TIGER is an acronym for Topographically Integrated Geographic Encoding and Referencing and an American
based format used by the United States Census Bureau to describe land attributes such as roads, buildings,
rivers, and lakes, as well as areas such as census tracts. The TIGER format forms a base for the US part of the
OpenStreetMap project.
4 The Integrated Transport Network, is part of the OS MasterMap and a format provided by the United Kingdom
governmental Ordnance Survey.
5 TeleAtlas is since 2008 wholly owned by navigation system company TomTom.
6 Navteq is since 2011 fully merged into NOKIA.
7 Web 2.0, is a term describing the state of the Internet as a collaboration focused information platform, where
the user produces content. The term is set against Web 1.0, where content was provided as ‚ready-to-use’ and no
interaction with the user was aimed (O’Reilly, 2005).
Proceedings of the 11th Space Syntax Symposium
150.5
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
In the context of space syntax analysis 2003, Thomson (2003) pioneered when proposing to
make use of street networks. His study focuses on theoretical and technical problems based
on the model construction rather than an investigation on how dierent models eects the
analysis. In the study, he highlights possibilities of generalizing road networks. Simultaneously
Dalton et al. propose to make use of TIGER data and present initial results of their analytical
work (2003). TIGER is a data format only used in the United States providing road-centre line
information among other geo-referenced spatial data. Dalton conducts a fractal analysis and
compares a TIGER dataset with a traditional hand-drawn axial map of Downtown Atlanta, US.
He highlights dierences in the results of both models and concludes that the result is caused
by the very dierent representation of space. While a long linear avenue with adjacent side
streets is represented by one long axial line in a traditional axial line map in the TIGER dataset
road centre-lines are segmented by nature and have a node at each intersection (this is the
case for any road centre-line map). Any topological investigation would thus lead to a highly
skewed outcome. Moreover, Dalton raises the theoretical problem of radii, emphasising the
need for a ‘relativisation’ due to the dierences within each system (ibid., p.9). While Dalton
did not propose a solution to the problem his argumentation led to a series of investigations by
Alasdair Turner.
In his study from 2005 Turner presents a methodology that overcomes this problem of
segmentation and ‘relativisation’ by drawing on advantages of space syntax applying ASA to
road centre-line maps in combination with a segment length weighted algorithm. The results
of his 2005 and 2007 study indicate that metric radii in combination with weighted choice
measures present not only a suitable alternative to SM models but, in fact, generate better
correlations with ow data in the tested case studies. Turner emphasises that his measure holds
congurational information while incorporating plausible cognitive and physical constraints
(2007, p. 553). Turner’s ndings are reasonable since road centre-line maps are fundamental
representations of the accessible – rights of way – movement network and incorporate more
detailed angular information than axial line models.
Dhanani et al. (2012) follow Turner’s ndings and conduct a comparative study of an axial line
model and two dierent types of road centre-line based models. As mentioned previously,
there are dierent sources for road centre-line maps. Dhanani et al. studies’ focus on two
very particular networks: the governmental ITN data set and the OSM VGI data. Their studies
aim to understand whether a VGI-based data set constitutes a reliable alternative compared
to governmental data sets in the light of space syntax analysis. Beside of Dalton’s (2003) and
Turner’s (2005, 2007, 2009) work, there are no other comprehensive studies where space
syntax measures are applied to governmental road centre-line data sets correlating results
with empirical data. This is surprising as both of the studies rely either on the American TIGER
data or the British Ordnance Survey data sets. The diculty here is that governmental road
centre-line maps are presented as a reliable and coherent source of data, yet, this is only true
for information within one data set 8 and very little is being said about their comparability in an
international context.
Dierences occur between governmental data sets not only on an international level but also
within countries. The British Ordnance Survey for example provides three dierent road centre-
line data products: the OS MasterMap layer Integrated Transport Network (ITN) layer, the OS
Open Roads layer and the Merdian 2 layer. All these data sets provide comprehensive road
network information and are designed for routing and road network analysis, yet, their level
of precision and coverage diers.9 This means that the total amount of nodes and coverage of
real world details such as roundabouts are not the same throughout the three data sets. More
importantly such data sets are not available in every country. Germany, Italy and France–to name
only some–do not provide freely accessible data sets. This is why, the question of comparability
needs to be answered and investigated for each country individually and alternative sources
8 It shall be noted that errors do occur in governmental data sets as well, but they usually follow a random
distribution.
9 See http://digimap.edina.ac.uk/webhelp/os/osdigimaphelp.htm#data_information/os_products/os_open_map_
local.htm for further information on the data sets and examples of their application.
Proceedings of the 11th Space Syntax Symposium
150.6
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
need to be found. The lack of comparable data makes it dicult for international comparative
approaches making use of such data sets, particularly in the context of space syntax.
2.2 ADVANTAGES AND DISADVANTAGES OF OSM DATA
In the light of this lack of comparable data, OSM data becomes more interesting as an
appropriate alternative to a segment map representation, which, in theory, provides a
comparable representation of space all over the world. OSM data is produced according to
a guideline indicating the level of precision and the handling of particular situations such as
divided highways, roundabouts, intersections or bridges (OpenStreetMap Wiki contributors,
2016). This makes the data, in theory, globally comparable. However, dierences in terms
of data quality arise due to the nature of its production and its contributors’ heterogeneous
understanding of street networks.
Understanding such dierences in quality is a non-trivial task in the realm of OSM data. There
is a set of ISO standardized quality measures to assess the quality of map-based VGI (OSM)
data. These measures are of particular interest for routing and navigation application, namely
positional accuracy and topological consistency (Senaratne et al., 2016, p. 6) and thus for a
space syntax application. Positional accuracy is a quantiable value reecting the dierence
between a mapped location and its real world location while topological consistency measures
how well topological relations (‘disjoin’, ‘meet’, ‘overlap’ or ‘equal’) are mapped. A simple
example for low positional accuracy would be a mapped intersection, of which the GIS location
is 20 meter further in the North than in reality. An example for bad topological consistency of
an intersection would be the case, in which two streets, which in reality are connected and
should share a common node, would not do so in GIS. To evaluate the two mentioned quality
measures it is necessary to compare the data set under investigation with the real world. This
is usually done by comparing the VGI data with ground-truth data. Ground-truth means data
that represents the respective exact location in reality. This is a theoretical value, rather than
an actually achievable goal for most GIS data sets. GPS systems feature on average a positional
accuracy of 6-10 metres to ground-truth. The ordnance survey MasterMap ITN data states its
positional accuracy with 1 metre in urban and 6 metres in rural areas against ground-truth.
Throughout the past decade, several authors have conducted comparisons of volunteered
geographic information with governmental as well as commercially produced geographic
information (Flanagin and Metzger, 2008; Neis et al., 2010; Zielstra and Zipf, 2010; Ludwig,
Voss and Krause-Traudes, 2011)10 to measure their quality. In the context of road centre-line
information the work by Mordechai Haklay was one of the rst to evaluate the quality of OSM
data (2010). Haklay used the British OS Merdian 2 road network as control measure to test
OSM data quality, his ndings indicated highest mapping qualities in urban and auent areas
and the lowest coverage in rural and poorer areas while positional accuracy ranges from over
70% to occasionally drop down to 20% (ibid., p.700). Overall OSM data covered 29% of England
based on a network from March 2008. In a subsequent study conducted in October 2009 this
percentage was already corrected to 65% of coverage (Haklay, 2009). This indicates a growth
of the network coverage by 36% within one year. Another study by Neis et al. (2011) dealing
with the case of Germany, compared the OSM network against the proprietary data set of
TomTom (formerly TeleAtlas) and estimated a complete coverage of the German OSM data by
the year of 2012. Moreover, already in 2011 the OSM data exceeded the topological consistency
and completeness of the TomTom network by 27% including pedestrian path ways (ibid.).
The continuous growth and its pace of the OSM data set, does not only make a coverage and
quality assessment dicult, but indicates that it is only a matter of time that full topological
consistency will be reached. The number of total users in the OSM community as well as their
nodal contribution to the network shows a growth of the total user number to 2,9 million since
the start of the project 2004 and gives insights in the pace of this process.
10 See Sehra et al. (2013) and Senaratne et al. (2016) for a comprehensive review of studies dealing with quality
assessment of VGI data.
Proceedings of the 11th Space Syntax Symposium
150.7
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 1 - Visualizing road updates. All roads shaded by how recently they have been updated by users.
Older imports are in green and blue, while cities with strong and active communities and the eect of
recent automated editing makes areas glow red. (2013) Source: https://www.mapbox.com/osm-data-report/
(retrieved on 1 August 2016)
Hakley et al. (2010, p. 11) investigate how many volunteers are needed to map an area thoroughly
concluding that areas mapped by more than 15 contributors per square kilometre feature a very
good positional accuracy of below 6 metres for resulting VGI data. In regard of the growing
numbers of contributors this leaves us to expect an equal rise in topographic consistency and
positional accuracy. An additional positive eect to the coverage of areas, beside the growing
number of contributors, is the fact that governmental agencies increasingly provide their
data for public usage. Likewise, are the American TIGER network as well as the AND Dutch
road network fully implemented in the OSM network, aiding not only to the coverage but
positional accuracy of the OSM data set. A visualised snapshot of the data and its topicality
reveals updating intervals, as well as showing that Great Britain and Germany are part of the
best-mapped countries of the OSM project (Figure 1). All of the above studies use ground-
truth data for the evaluation of VGI quality. Still, such data is not available in every country and
more diculties for the assessment of VGI data arise due to the lack of ground-truth data for
comparison (Senaratne et al., 2016, p. 6). To overcome this lack of ground-truth data, Keßler
and de Groot (2013) propose a method to indicate quality of VGI via trust assessment models.
Their approach is based on a trust assessment model of the independent contributions in an
OSM data set. Albeit presenting promising results, the methodology is at an early stage of
development and does not propose an applicable method for the eld. At the present stage,
this leaves the research with as-good-as complete network for some countries with reasonably
accurate precision, but a manual control of the entire data set by the researcher stays a
necessity. With regard to future research the OSM will very likely constitute the most coherent
freely available data set.
Dhanani et al. (2012, p. 30), assess the usage of OSM in space syntax to be problematic and
describe the data as lacking ‘of consistency [,…] accuracy and coverage’. Their study calls on
researcher to rely on governmental data such as the British OS MasterMap ITN, yet, as mentioned
earlier, as data is not accessible in every country and level of detail diers throughout dierent
data sets, this approach remains unsatisfactory: The OS MasterMap ITN network covers only
the vehicular network disregarding any path or street that is only accessible to pedestrians.
The resulting vehicular centred spatial representation can therefore only be used to evaluate
vehicular structures. Space syntax segment map representation on the other hand sees space
through the eye of an individual moving in space and constitutes a sharp contrast to a vehicular
Proceedings of the 11th Space Syntax Symposium
150.8
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
only street network. There are also other diculties within the ITN data set that render an ad
hoc use impossible. Dhanani et al. note that the ITN network comprises all trac management
features including trac islands, articial cul-de-sacs or roundabouts (ibid., p.6). According
to the authors, using such data creates a ‘disjoint and fragmented network’ particularly if a
researcher is interested in other modes than a purely vehicular estimation. The usage of such
data is not recommendable without any prior processing. Prior processing is also necessary for
OSM data making it indispensable to develop a strategy to overcome said inconsistency and
arrive at a comparable network for any given case.
3. OSM DATA STRUCTURES AND GIS SIMPLIFICATION PROCESSES
The following section gives an overview of the necessary components to create a road network
based on OpenStreetMap data and the necessary steps of post processes to allow an application
in space syntax ASA.
At present, OSM data sets are divided into four dierent elements: nodes, lines, surfaces
and relations. For an ASA only line information is necessary, but not all of the available line
information and categories are useful. The OSM wiki provides extensive accounts on all dierent
key categories and their morphology (OpenStreetMap Wiki contributors, 2017), it is important
for each researcher working with OSM data to make him/herself familiar with all categories and
morphologies. Decisions about which category to exclude might dier for example in cities in
developing countries. The following steps should to be considered as a general guidance: For
the purpose of network analysis only components with the key highway=* shall be used. This
key denes any kind of road, street or path and their respective importance in the network
hierarchy (from the most important ‘motorway’ to the least ‘service’) and, thus, gives a good
account of the rights of way network. The following list assess which are recommendable to be
included in a network for an application in ASA: highway=motorway; trunk; primary; secondary;
tertiary; unclassied; residential; motorway_link; trunk_link; primary_link; secondary_link;
tertiary_link; living_street; pedestrian (ibid.). Particular care needs to be taken with the key
pedestrian as it includes pseudo polyline information of squares and these need to be cleaned
and subsequently broken into individual segments. Other sub keys such as highway=service;
path or bridleways can be included but are not recommended, as they are of very small scale
and might otherwise be eradicated in a subsequent simplication process.
With a view to this selected data there are three main diculties that occur when applied in a
space syntax context.
1. Topological inconsistency occurs if street segments are supposed to share a connecting
node but due to positional inaccuracy fail to do so. This is often the case at intersections
of dierent contributors. Even a small gap between two nodal ends of 1 cm can create
a network fragmentation. It is, therefore, necessary to process and clean the data from
these inconsistencies.
2. Trac management components are network details that are necessary for vehicular
trac management but have no immediate impact on cognitive route decision-making.
Such details are for example roundabouts, small trac islands or motorway trunks.
Ideally roundabouts are simplied into simple intersections whereas meandering trunk
links are represented by single links. Moreover, this is also the case with regard to dual
line representations. Space syntax analysis is a non-directional approach in the sense that
the possible travel directions are not taken into consideration and each space is treated
as equally accessible. A dual line representation constitutes only a reasonable option if
directions are taken into consideration. Hence, the model needs to be cleaned from said
dual line representations.
3. Redundant or excessive nodal information are often problematic when using OSM data.
Although the OSM guide notes that nodes should be used in an economic manner,
contributors often have dierent interpretations of what ‘economic’ means. This is
particularly the case for curved roads, but also occurs on straight lines. Ideally each street
is simplied to its fundamental segment.
Proceedings of the 11th Space Syntax Symposium
150.9
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
In order to overcome these diculties a series of GIS algorithms have been developed. The
following proposed solutions are employing the GIS software ArcGIS Desktop 10.2 from
Esri. I employ ArcGIS because it is the only software that provides solutions for all three said
diculties. At present, only a few of the solutions presented here can be achieved with open
source GIS software packages. Due to the scope of this paper only a brief description of the
applied core functionalities will be given. Figure 2 shows a workow diagram for the proposed
solutions, while Figure 3 gives an illustration of each obstacle and its favoured solution after the
application of the simplication method presented here.
Figure 2 - Workow of ArcGIS tools and algorithms to solve: 1. topological inconsistency; 2a. dual line re-
moval; 2b. road detail removal and 3. line simplication.
Proceedings of the 11th Space Syntax Symposium
150.10
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 3 - Illustration of each diculty found in OSM data: 1. topological inconsistency; 2a.
dual line removal; 2b. road detail removal and 3. line simplication as well as the condition
after application of the simplication method.
1. Starting with the approach of solving topological inconsistency (Figure 2:1), it should
be mentioned that a lack of network information, such as entire missing streets cannot
be solved through automated processing and that the OSM data needs to be carefully
checked by the research prior to any post-production. More so, this is a strategy to
overcome small inconsistencies that are dicult to identify manually. The proposed
process reconnects topological inconsistencies by a given tolerance distance and in
a subsequent step merge segments that can be considered as independent streets
(from intersection to intersection) together. This will leave the researcher with a street
network of real segments and consistent topological information. The two core ArcGIS
functionalities the workow is based on are ‘integrate’ and ‘unsplit’.
Proceedings of the 11th Space Syntax Symposium
150.11
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
The integrate tool is applied to extracted nodal information, rather than the actual
line information, to overcome misalignment at intersections. Integrate maintains the
integrity of shared nodal feature information by making features coincident if they fall
within the specied x, y tolerance. Features that are considered identical or coincident
are merged. In a subsequent step the newly generated nodal point information is used as
a basis for a snap command of the initial street network. This will consequently connect
lines, which feature topological inconsistencies, at a new point based on the location of
their nodal line ends.
The unsplit tool is then applied to the now topological consistent line network. The aim
is here to aggregate single part line features into multipart features in order to arrive
with continuous street segments. Unsplit merges lines that have coincident endpoints.
This can be done by relying on any given attribute information or, as in this case, solely
by geometric relationships. Merged lines are of particular importance with regards to
further simplication processes.
2. The next diculty is the existence of trac management details and dual line
representations in the data sets (Figure 2:2a & 2b). Not only do such details (roundabouts,
trac islands, etc.) create dierences in angular movement, while the general journey
direction stays the same, but more importantly they increase the total number of journeys
(dual line highways) and skew analytical results towards an emphasis of such details.
Especially in the light of none directed centrality analysis dual lines make little sense. This
could be negligible if trac management details were normally distributed throughout
the street network. However, this is, not the case with most examples and particularly
not with inter-city and regional scales. There are four main ArcGIS components, ‘merge
divided roads’, ‘collapse dual lines’, ‘collapse road details’ and ‘integrate’ that help to
remove such dual lines and reduce low-level street network complexity.
The merge divided roads is an algorithm that merges road segments, which are parallel
along a signicant distance into a single centre line. The merging process is based on
common attributes that can be computed on the basis of the initial highway keys. It is
fundamental that the merge eld parameters are established properly to avoid conicts
during the process. The divided roads algorithm can be applied to entire data sets and
maintains topological relations with adjacent streets.
The collapse dual lines to road centerline is an algorithm designed to derive with centre
lines from a base of street perimeters. It is, therefore, a less sophisticated form of
simplication and it is not recommended to perform the algorithm on large datasets
including multiple-lane highways with interchanges, ramps, overpasses and underpasses.
In individual cases where the merge divided roads tool does not arrive with satisfactory
results, the collapse dual line to road centerline tool can form a useful alternative.
The collapse road detail, on the other hand is an algorithm that depicts small road segment
details and open congurations that interrupt the general trend of a road network and
collapses or replaces them with a simplied feature. The collapse distance on which the
tool performs is dened by the maximum size of the largest road detail and can dier for
each model. If the collapse road detail tool does not solve or remove some of the details
the integrate tool explained earlier constitutes an appropriate alternative. Particular care
needs to be taken when using integrate on road details as it can impact the topological
consistency of the data and should hence not be performed on entire data sets but single
cases.
3. Line simplication is usually applied when segment records feature far more data than
necessary for computer analysis or visual representations (Figure 2:3). In the case of
space syntax and the use of VGI street networks this poses a conceptual question aside
of excessive data. While road-centre lines depict the centre of the road an axial line
(as base for a segment map line) is based on the longest line of sight. A generic street
usually features a much larger eld of vision than that of a single line. While axial lines
fundamentally connect convex spaces these lines naturally pervade more than one space
Proceedings of the 11th Space Syntax Symposium
150.12
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
at once. Road-centre lines on the other hand simply represent the centre of the road
and, therefore, feature excessive angular information that does not impact the eld of
vision or accessibility and, thus, has no eect on the actual movement in space. This is
why, a removal of such road details should be based on the eld of vision of each street,
i.e. the street width. Since road-centre lines give a precise account of the centre of each
street segment a simplication process should allow the newly generated feature to
deviate to at least the extent of the eld of vision. Such processes can be performed by
the Douglas-Peuker Algorithm (DPA) (1973). The DPA is broadly considered to deliver the
best perceptual representations of the original segment and generates new segments
based on a deviation tolerance. In ArcGIS this can be done by applying the simplify line
tool.
The simplify line tool reduces and removes redundant nodes of line features. Among
others, when applied with the POINT_REMOVAL functionality it employs the DPA. The
aim of the algorithm is to extract the essential segment form based on a previously
selected o set tolerance. The strength of the algorithm is its reproducibility and process
speed, and that it arrives at the same solution to the same given problem.
If the above steps of the methodology are followed the simplied version of a road-
centre line map (SIMP) looks visually as well as topologically much closer to an axial line
representation.
4. MODEL EVALUATION METHODOLOGY
In order to test if the theoretically laid out version of a simplied OSM network (SIMP) constitutes
a comparable alternative to a segmented axial line map and is, thus, suitable for the purpose
of analysis of dierent scales and very large ones in particular the model will be analysed and
correlated with results from an ASA of a segment map, ITN and OSM model. The comparison
extends and the builds on methodologies by Eisenberg (2007), Turner (2007) and Dhanani et al.
(2012).
Eisenberg (2007, p. 5) focused on comparison of dierent axial line models for the same cities.
The dierent models that Eisenberg compares are developed as a by-product of variations
in analytical scales (pedestrian, bicycle and vehicular) and variations in the detail of the base
information used for the production of the axial line maps. Eisenberg highlights that three
indicators are of interest for a comparison. First, the impact of base map scales; Second,
dierent levels of detail; And, third, dierent city morphologies (ibid., p.5). All aspects are
directly transferable to the dierent network models previously introduced. Eisenberg’s
ndings suggest that the analysis should focus on ‘rank correlation measures’ in order to have
a meaningful comparison (ibid., p.8). Eisenberg’s ‘rank correlation measures’, are applicable
to every kind of network representation. This measure simply compares values and their
respective rank within the data set. With Eisenberg’s measure an appropriate method for the
aimed analysis is established where numbers of lines dier signicantly and the resulting values
do not form a comparable unit.
In addition to ‘rank correlation’ this comparison will draw on the methodology of Turner (2007).
Turner proposed an angular based analysis in combination with segment length-weighting and
the introduction of a metric length based radius. While an angular based analysis incorporates
the cognitive dimension of route choices, the reasoning behind a segment length-weighting is
to overcome the large dierences in segment numbers between the dierent representations
(ibid., p.541). Turner shows how his propositions are an advancement for space syntax analysis
in general and in the context of road-centre line networks in particular.
Finally the above proposed methods will be merged with a methodology by Dhanani et al.
(2012). Dhanani et al. conducted a comparison of road-centre line networks against axial line
models using a general description of the network characteristics followed by a topological
and metric step depth analysis from the most central segment. Although the outcome of the
topological step depth showed interesting results the application of topology on a road-centre
Proceedings of the 11th Space Syntax Symposium
150.13
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
line network remains inappropriate as road-centre lines topological information is highly
skewed by its nodal information. The measure of topology in space syntax analysis is based
on the cognitive and visual space in the sense that what is considered as one space in space
syntax would result in several spaces in a road-centre line network. The analysis will only draw
on the measure of metric step depth (MSD) for comparisons as MSD is not aected by nodal
information.
In summary, the following comparison is based on four dierent road network models of the
centre of the city of Leeds. The city of Leeds was selected because it features a variety of dierent
network details such as motorways, trac management details as well as local paths. The road
network models are: the Ordnance Survey ITN network, the OSM network, a simplied version
of the OSM (SIMP) and a segmented axial line model (SM). The ITN network and the OSM
data are not simplied but instead used as they are provided by the organisations. Moreover,
the ITN and OSM networks where controlled on topological consistence, yet, no irregularities
were found. Some network categories, as those mentioned in the OSM data sections, have
been removed from the OSM data set while trac management details remained unchanged.
The four models are compared in regard to their network characteristics and analysed on
14 dierent radii from 100 up to the entire system n11 using angular segment analysis with
segment length weighting. The models are analysed on closeness and betweenness centrality.
The resulting structures of three exemplary scales are visually compared. Then, subsequent
correlations are conducted using ‘rank correlation measures’. To facilitate comparisons mean
values of coincident segments of the ITN, OSM and SIMP with the SM model are plotted on
each respective SM segment.
4.1 RESULTS
11 The applied scales are: 100, 150, 200, 300, 500, 800, 1300, 1800, 2500, 3200, 4100, 5000, 6100 and n.
Figure 4 - Detailed section of dierent network models of ITN, OSM, SIMP and SM.
Proceedings of the 11th Space Syntax Symposium
150.14
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 4 shows a small section of each of the modelled areas. The section of the ITN network
shows trac islands as well as road interruptions. Some roads have signicant angular turns
just before their connection with the adjacent road. This is because for trac management
purposes rectangularity is preferred. In the light of angular segment analysis, Dhanani et al.
(ibid., p.10) consider this preference an important aspect and the most detailed and ‘optimal’
account of the street network. The aerial photo of the area (Figure 4) shows that at this point a
straight connection is a more reasonable account of the real world situation. Additionally, at the
lower right there is a road divergence into two separate lanes. A noteworthy detail is also that
roads, which could be considered as intersecting in reality do not share a common node in the
road network, due to a 5-10 metre distance of their road-centre.
Statistics ITN OSM Axial SIMP
Segments 15049 9308 5072 3908
Total length (m) 283410 276388 240534 238848
Computation time (min) 14.31 4.49 1.21 0.44
Table 1 - Network characteristics for each model.
Table 1 highlights the network characteristics for the four models and how they dier
numerically. The ITN network features the longest total network length with 283410 metres. This
is particularly due to the several roundabouts and trac management details within the model.
The comparison of trac management details with the length of the ITN and OSM networks
enables a rough account of the eect on the length of the network. This account does not come
to its fullest as the OSM network features streets and connections that are not represented in
the ITN. The several multi line motorway roads, which are represented by a single segment in a
segmented axial line and SIMP model cause a large dierence of 40km of the ITN and OSM data
in comparison to the segmented axial model. Comparing all networks, the dierence in number
of segments is striking. The ITN model has three times more segments than the segment map
representation. This dierence is due to the curved roads and roundabouts, which feature large
numbers of segments in order to give precise accounts on the length of the lines. While this
exemplies the detailed account on angular changes in road centre-line networks, it also shows
the inherent problem this data has when it comes to space syntax analysis. The computational
time is O(n2) to the number of segments. Generally speaking, the ITN and OSM are similar
in their measures and the dierence in number of segments is as expected. With regard to
the segmented axial line and SIMP model the question is whether the SIMP model, with 33%
less segments, does also stores less information. The number dierences can be explained
by the ‘cleaning’ of intersecting spaces: Whenever three segments intersect with each other
segmented axial line models tend to create clusters of very short segments. Additionally, when
the axial line model is converted to a segmented axial line, stubs that fall over 40% of the line
length are not removed and might also contribute to this dierence. The SIMP model features
almost the same length as the segmented axial line model pointing towards a similar degree of
spatial representation.
Proceedings of the 11th Space Syntax Symposium
150.15
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
These observations become more apparent with a look at the histograms for segment length
distribution for each m odel type. While the ITN network exhibits an even increase of segment
length with declining frequency, the OSM shows an initial increase indicating fewer stubs and
curve segmentation than the ITN. Moreover, the short line cluster eect of the SM model
becomes visible with almost thousand segments in the range of approximately 1-10 metres.
Contrarily, the SIMP model has a steep increase of frequency with a peak at a mid range of
approximately 30 metres indicating less of a short line information. The simplication range
used during the simplication process has an inuence on this peak.
Dhanani et al.’s (ibid., p.25) study shows that dierences between road centre-line network and
axial line models are consistent in their appearance and concludes that the dierent models do
not form a fundamentally dierent structure of the spatial conguration. In the next step I will
compare the new SIMP model with this assumption. Figure 6 shows the number of segments
for nine dierent radii where the maximum is 2,5km as this is the distance at which the entire
system was captured (in other words n). For the four models, the total number of segments
reached per metric distance increases in relation to the total number of segments. The semi-log
plot highlights these similarities and dierences, especially at lower scales. The SM and SIMP
model, exhibit a similar development, while the OSM and ITN, which were initially similar,
disperse towards growing metric distances and due to the increase of network details. Unlike
the values for the central segment the curve for the edge segment shows a slightly uneven
development. This becomes clearer in the semi-log plot of the data. Here, particularly the
development around the scale of 500 metres unveils that there are underlying dierences in
the complexity of the models that might have an eect on the analysis.
Figure 5 - Histogram of segment length distribution of each of the four models.
Proceedings of the 11th Space Syntax Symposium
150.16
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
In order to arrive at a better and more detailed account of the impact of dierences in the
network morphologies, I conduct a comparison of betweenness and closeness centralities
using a segment angular analysis with segment length weighting. The models are analysed on
14 dierent radii. The applied scales are; 100, 150, 200, 300, 500, 800, 1300, 1800, 2500, 3200,
4100, 5000, 6100 and n. Two of these scales, 800 and n, are visualised in order to understand the
geographic distribution of dierences. Figure 7 shows the results for betweenness centrality.
Figure 8 shows the results for closeness centrality. The values of each gure are broken down
using a quantile division. This is done to overcome signicant outliers in the data sets that make
a natural break highly skewed and the resulting maps illegible. These circumstances make
it necessary to process the data in a GIS programme rather than applying the implemented
symbologies of depthmapX.
Figure 6 - 1a: Number of segments for dierent metric step depth from the most central segment for ITN,
OSM, SM and SIMP models. 1b: Semi-log plot of the same data set. 1c: Number of segments for dierent
metric step depth from an edge segment for ITN, OSM, SM and SIMP models. 1d: Semi-log plot of the same
data set.
Proceedings of the 11th Space Syntax Symposium
150.17
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 7 - ITN, OSM, SIMP, and SM models analysed on ASA SLW betweenness centrality on radius metric 800
(1) and radius n (2).
Proceedings of the 11th Space Syntax Symposium
150.18
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 8 - ITN, OSM, SIMP, and SM models analysed on ASA closeness centrality on radius metric 800 (1) and
radius n (2).
Proceedings of the 11th Space Syntax Symposium
150.19
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
The results show that all models exhibit comparable patterns on all of the two visualised scales
and both measures of betweenness and closeness centrality. This conrms the initial ndings of
Dhanani et al. (2012). However, similarities in the results were much stronger between the OSM
network and the SM than they were between ITN and SM. Nominal segment dierences appear
to have a higher impact on betweenness centrality than on closeness centrality. Models with
large numbers of short segments and high degree of precision, such as the ITN network, are,
thus, more likely to be aected by outliers and unexpected clusters, than models with fewer
short segments. Moreover, the ITN network shows high values on all scales in the motorway
network. The SIMP model showed patterns that were visually stronger related to the SM model
than to the ITN or OSM and more similar to the OSM compared with the ITN. This is rather
unexpected as SM models are thought to be intrinsically dierent.
After getting an understanding of dierences and similarities in the geographical distribution
of the data between the dierent models and the SIMP model in particular a nal analysis of the
statistical extent of these observations is conducted. This will give an account of how models
behave in comparison to each other across all scales. As elaborated before, the analysis draws
on Eisenberg’s proposed ‘rank correlation measure’. To give a more detailed account, dierently
to Eisenberg, this analysis will compare all segments that are intersecting rather than only 10%
of highest values proposed by Eisenberg (2007). This is done by plotting mean values of the ITN,
OSM and SIMP on the SM model. The SM model is used as a base and comparisons are only
conducted with streets whose middle point falls into a 10-metre distance of a SM segment.
These middle points are then snapped to the closest segment and plotted on the SM model.
If more than one street segment of an ITN, OSM or SIMP model falls into this category, their
mean is calculated and plotted on the SM model instead.
Eisenberg’s rank correlation is based on Spearman’s Rank correlation (ibid.). Spearman’s Rank
correlation coecient is generally used to identify and test the strength of a relationship
between two sets of data. It tests if the relationship of both variables can be described by a
monotonic function. Ideally, the SIMP model could predict the segmented axial line model by
such monotonic function. In addition to this, a Pearson correlation will be conducted. Rather
than correlating the dierent ranks of each variable, a Pearson correlation works with the
actual values of the variables and measures their linear correlation. Both correlations provide a
coecient R2 indicating how related the variables are with each other. A coecient of 1 indicates
that the two models are identical. Any value below 1 describes the degree of dierence. One
can hence compare the dierences between all models statistically and provide a correlation
coecient to describe the tness of the SIMP model for the purpose of space syntax ASA. The
analysis is based on 14 dierent scales for both space syntax measures of betweenness and
closeness centrality. Figure 9 and Figure 10, show Pearson and Spearman correlations of ITN,
OSM and SIMP compared with the segmented axial model and, subsequently, the same for all
models correlated against the SIMP model.
Starting with Figure 9 the ndings from the initial visual description becomes also statistically
apparent. A rst observation is that the Spearman rank correlation provides more consistent
results across scales and measures with weaker dierences and higher scores. The Pearson
correlation on the other hand shows much stronger dierences in the four data sets but features
a signicant outlier on the scale of 100 metres for closeness centrality. In regard to the single
models the ITN model shows lower correlations across both Pearson and Spearman measures
and on both betweenness and closeness centrality. Particularly interesting is the signicant
drop towards higher radii, with a lowest correlation of 0,56 on Pearson for betweenness and
closeness. This increases at the Spearman’s rank, however, the general tendency towards
lower correlation at higher radii persist. In terms of the visual observations made earlier this
is caused by trac details and the strong representation of motorway features. The OSM and
SIMP model on the other hand show very comparable correlation developments. An exception
of this is the Persons correlation for betweenness centrality of the OSM model where similar
to the ITN a sudden drop at higher radii is visible. The SIMP model correlates stronger across
all measures with the highest scores of 0,983 for Spearman correlations of closeness centrality
metric 1300 and 0,919 for betweenness centrality. Contrary to OSM and ITN the correlations for
SIMP are very consistent.
Proceedings of the 11th Space Syntax Symposium
150.20
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 10 shows the Pearson and Spearman correlations for 14 dierent scales and closeness
and betweenness centralities. However, this time ITN, OSM and SM models are compared with
SIMP. The general correlation developments are very similar to the ones we have observed
previously, with a progressive drop of values towards higher radii. Interesting is at this point how
ITN and OSM behave compared to the SIMP model. While the ITN networks shows a slightly
weaker correlation, the OSM correlates much stronger. This was on one hand an expected
result, as the SIMP model is entirely based on the OSM. On the other hand in the light of the
overall comparison it seems as if the simplication process brought the simplied OSM model
much closer to the segmented axial line representation than expected.
Figure 9 - 1: R2 of a Pearson correlation for ASA segment length weighted betweenness centralities
(1a) and closeness centrality (1b) for 14 dierent metric radii (from 100 metres to n) for the three
dierent network models SIMP, OSM and ITN against the SM model. 2: R2 of a Spearman correlation
for ASA segment length weighted betweenness centralities (2a) and closeness centrality (2b) for 14
dierent metric radii (from 100 metres to n) for the three dierent network models SIMP, OSM and
ITN against the SM model (left). Correlation is signicant at the 0.01 level (2-tailed), N=3172.
Proceedings of the 11th Space Syntax Symposium
150.21
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 10 - 3: R2 of a Pearson correlation for segment length weighted betweenness centralities (3a)
and closeness centrality (3b) for 14 dierent metric radii (from 100 metres to n) for the three dierent
network models SM, OSM and ITN against the SIMP model (left). 4: R2 of a Spearman correlation for
segment length weighted betweenness centralities (4a) and closeness centrality (4b) for 14 dierent
metric radii (from 100 metres to n) for the three dierent network models SM, OSM and ITN against the
SIMP model (left). Correlation is signicant at the 0.01 level (2-tailed), N=3172.
These dierences become more apparent with regard to a log-log scatterplot of betweenness
and closeness centrality of the global scale n (Figure 11). The diagram shows a log-log scatterplot
of each of the measures allowing a visual comparison of outlier distribution within each data
set. The more dispersed the values are the less they correlate while linear consolidation implies
stronger correlations. This is clearly visible for the log-log plot of axial and SIMP while both
other models show stronger dispersion. The ITN model shows outliers across the values from
low to high, which is particularly the case for closeness centrality. To summarize, the results
Proceedings of the 11th Space Syntax Symposium
150.22
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Figure 11 - Log-Log plots for the SM model compared to ITN, OSM and SIMP respectively for ASA SLW
betweenness and ASA closeness centralities on radius n.
Proceedings of the 11th Space Syntax Symposium
150.23
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
show that the four models dier especially in terms of the number of short length segments.
This dierence can be described by an exponential relation and has a signicant impact on
the computational time needed for the analysis. The results of the metric step depth analysis
conrm the ndings of Dhanani et al. (2012) and show that all models share a similar complexity
in terms of their nodal distribution. However, the analytical space syntax analysis showed that,
albeit, there is a similar distribution in the data in general the geographic location of these
dierences has an impact on the results. The ITN network is strongly inuenced by its emphasis
on vehicular movement and trac management details. This makes it less comparable to the
segmented axial line model than the OSM model or the SIMP.
5. CONCLUSIONS
Concluding, this paper elaborated the tness of OSM data in space syntax analysis, it proposed
an ArcGIS simplication workow and presented the theoretical reasoning behind the method.
The nal tness tests showed that the simplied OSM network (SIMP) exhibits very strong
similarities with the traditional segmented axial line model across all investigated cases. It
features the topological and angular information of the OSM network with the simplistic
representation of a segmented axial line model. This is rather surprising, because the alterations
in the model are mainly based on segment nodal reduction and minor topological alteration.
The Pearson and Spearman correlation analysis showed that the SIMP model is in fact stronger
related to the segmented axial model than to the OSM model. The strong similarity between
SIMP and segmented axial also poses question to weather axial line models are such intrinsically
dierent representations.
Overall the ndings suggest that a simplied OSM network forms an appropriate model for
space syntax analysis, particularly in the light of regional investigations where the production
of an axial line model is not a feasible option.
Proceedings of the 11th Space Syntax Symposium
150.24
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
REFERENCES
Batty, M. and Rana, S. (2004) ‘The automatic denition and generation of axial lines and axial maps’, Environment
and Planning B: Planning and Design, 31(4), pp. 615–640. doi: 10.1068/b2985.
Carranza, P. M. and Koch, D. (2013) ‘A Computational Method for Generating Convex Maps’, in Kim, Y. O., Park, H. T.,
and Seo, K. W. (eds) Proceedings of the Ninth International Space Syntax Symposium. Seoul: Sejong University,
p. 064: 1-11. Available at: http://www.diva-portal.org/smash/get/diva2:679788/FULLTEXT01.pdf.
Dalton, N. S., Peponis, J. and Dalton, R. (2003) ‘To tame a TIGER one has to know its nature: Extending weighted
angular integration to the description of GIS road-centerline data for large scale urban analysis’, 4th International
Space Syntax Symposium, pp. 1–10.
Desyllas, J. and Elspeth, D. (2001) ‘Axial Maps and Visibility Graph Analysis. A comparison of their methodology and
use in models of urban pedestrian movement’, in Proceedings of the Third International Space Syntax Symposium.
Atlanta, USA, p. 27.1-13. Available at: www.ucl.ac.uk/bartlett/3sss/papers.../27_desyllas.pdf.
Dhanani, A., Vaughan, L., Ellul, C. and Griths, S. (2012) ‘From the axial line to the walked line: Evaluating the utility
of commercial and user-generated street network datasets in space syntax analysis.’, Proceedings of the Eighth
International Space Syntax Symposium. Santiago de Chile: PUC, pp. 1–32. Available at: http://discovery.ucl.
ac.uk/1308812/.
Douglas, D. H. and Peucker, T. K. (1973) ‘Algorithms for the Reduction of the Number of Points Required for Represent
a Digitzed Line or its Caricature’, Cartographica (The Canadian Cartographer), 10(2), pp. 112–122.
Eisenberg, B. (2007) ‘Calibrating axial line maps’, in Proceedings of the Sixth International Space Syntax Symposium.
Istanbul, Turkey, pp. 1–14.
Flanagin, A. J. and Metzger, M. J. (2008) ‘The credibility of volunteered geographic information’, GeoJournal, 72(3–4),
pp. 137–148. doi: 10.1007/s10708-008-9188-y.
Goodchild, M. F. (2007) ‘Citizens as sensors: The world of volunteered geography’, GeoJournal, 69(4), pp. 211–221.
doi: 10.1007/s10708-007-9111-y.
Haklay, M. (2009) OpenStreetMap and Ordnance Survey Meridian 2 – Progress maps. Available at: https://povesham.
wordpress.com/2009/11/14/openstreetmap-and-ordnance-survey-meridian-2-progress-maps/ (Accessed: 10
June 2016).
Haklay, M. (2010) ‘How good is volunteered geographical information? A comparative study of OpenStreetMap and
ordnance survey datasets’, Environment and Planning B: Planning and Design, 37(4), pp. 682–703. doi: 10.1068/
b35097.
Haklay, M., Basiouka, S., Antoniou, V. and Ather, A. (2010) ‘How Many Volunteers Does it Take to Map an Area Well?
The Validity of Linus’ Law to Volunteered Geographic Information’, The Cartographic Journal, 47(4), pp. 315–322.
doi: 10.1179/000870410X12911304958827.
Hanna, S., Serras, J. and Varoudis, T. (2013) ‘Measuring the structure of global transportation networks’, in Kim, Y. O.,
Park, H.-T., and Seo, K. W. (eds) Ninth International Space Syntax Symposium. Seoul: Sejong University.
Hillier, B. and Hanson, J. (1984) The Social Logic of Space. Cambridge: Cambridge University Press.
Hillier, B. and Iida, S. (2005) ‘Network and psychological eects: a theory of urban movement’, (1987), pp. 475–490.
Available at: papers2://publication/uuid/51712050-C088-4BAE-B7A0-6A4372B28C46.
Hillier, B., Penn, A., Hanson, J., Grajewski, T. and Xu, J. (1993) ‘Natural Movement: or, conguration and attraction in
urban pedestrian movement’, Environment and Planning B: Planning and Design, 20, pp. 29–66.
Keßler, C. and de Groot, R. T. A. (2013) ‘Trust as a Proxy Measure for the Quality of Volunteered Geographic
Information in the Case of OpenStreetMap’, in Lecture Notes in Geoinformation and Cartography, pp. 21–37. doi:
10.1007/978-3-319-00615-4_2.
Law, S. and Versluis, L. (2015) ‘How do UK regional commuting ows relate to spatial conguration ?’, in Karimi, K.,
Vaughan, L., Sailer, K., Palaiologou, G., and Bolton, T. (eds) Proceedings of the 10th International Space Syntax
Symposium. London: Space Syntax Laboratory, The Bartlett School of Architecture, University College London,
p. 74:1-74:21.
Ludwig, I., Voss, A. and Krause-Traudes, M. (2011) ‘A Comparison of the Street Networks of Navteq and OSM in
Germany’, in Geertman, S., Reinhardt, W., and Toppen, F. (eds) Berlin, Heidelberg: Springer Berlin Heidelberg
(Lecture Notes in Geoinformation and Cartography), pp. 65–84. doi: 10.1007/978-3-642-19789-5_4.
Proceedings of the 11th Space Syntax Symposium
150.25
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Neis, P., Zielstra, D., Zip, A. and Strunck, A. (2010) ‘Empirische Untersuchungen zur Datenqualität von OpenStreetMap
- Erfahrungen aus zwei Jahren Betrieb mehrerer OSM-Online-Dienste’, in Angewandte Geoinformatik 2010.
Salzburg, Austria, p. 6. Available at: http://www.vde-verlag.de/proceedings-en/537495055.html.
Neis, P., Zielstra, D. and Zipf, A. (2011) ‘The Street Network Evolution of Crowdsourced Maps: OpenStreetMap in
Germany 2007–2011’, Future Internet, 4(1), pp. 1–21. doi: 10.3390/4010001.
O’Reilly, T. (2005) What Is Web 2.0: Design Patterns and Business Models for the Next Generation of Software, O’Reilly
Media, Inc. Available at: http://www.oreilly.com/pub/a//web2/archive/what-is-web-20.html (Accessed: 1 March
2016).
OpenStreetMap Wiki contributors (2016) Editing Standards and Conventions, OpenStreetMap Wiki. Available at:
http://wiki.openstreetmap.org/wiki/Editing_Standards_and_Conventions (Accessed: 3 June 2016).
OpenStreetMap Wiki contributors (2017) Key:highway., OpenStreetMap Wiki. Available at: http://wiki.openstreetmap.
org/w/index.php?title=Key:highway&oldid=1332527 (Accessed: 30 January 2017).
Penn, A., Conroy, R., Dalton, N., Dekker, L., Mottram, C. and Turner, A. (1997) ‘Intelligent Architecture: New Tools for
the Three Dimensional Analysis of Space and Built Form’, in Proceedings of the First International Space Syntax
Symposium. London: The Bartlett School of Architecture, p. 30.1-19.
Penn, A., Hillier, B., Banister, D. and Xu, J. (1998) ‘Congurational modelling of urban movement networks’,
Environment and Planning B: Planning and Design, 25(1), pp. 59–84. doi: 10.1068/b250059.
Peponis, J., Wineman, J., Bafna, S., Rashid, M. and Kim, S. H. (1998) ‘On the generation of linear representations
of spatial conguration’, Environment and Planning B: Planning and Design, 25(4), pp. 559–576. doi: 10.1068/
b250559.
Peponis, J., Wineman, J., Rashid, M., Kim, S. H. and Bafna, S. (1997) ‘On the description of shape and spatial
conguration inside buildings: convex partitions and their local properties’, Environment and Planning B, 24,
pp. 761–781.
Ratti, C. (2004) ‘Space syntax:some inconsistencies’, Environment and Planning B: Planning and Design, 31, pp. 487–
499.
Sehra, S. S., Singh, J. and Rai, H. S. (2013) ‘Assessment of OpenStreetMap Data - A Review’, 76(16), pp. 17–20. doi:
10.5120/13331-0888 10.5120/13331-0888.
Senaratne, H., Mobasheri, A., Ali, A. L., Capineri, C. and Haklay, M. (Muki) (2016) ‘A review of volunteered geographic
information quality assessment methods’, International Journal of Geographical Information Science,
8816(August), pp. 1–29. doi: 10.1080/13658816.2016.1189556.
Serra, M., Hillier, B. and Karimi, K. (2015) ‘Exploring countrywide spatial systems: Spatio-structural correlates at
the regional and national scales’, in Karimi, K., Vaughan, L., Sailer, K., Palaiologou, G., and Bolton, T. (eds)
Proceedings of the 10th International Space Syntax Symposium. London: Space Syntax Laboratory, The Bartlett
School of Architecture, University College London, p. 84.1-84.18.
Thomson, R. (2003) ‘Bending the axial line: smoothly continuous road centre-line segments as a basis for road
network analysis’, in Proceedings of the Fourth International Space Syntax Symposium. London, p. 10. Available
at: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.200.8956.
Turner, A. (2001) ‘Angular Analysis’, in Proceedings of the Third International Space Syntax Symposium. Atlanta, p.
30.1-30.11. Available at: http://discovery.ucl.ac.uk/35952/.
Turner, A. (2005) ‘Could A Road-centre Line Be An Axial Line In Disguise?’, Proceedings 5th International Space Syntax
Symposium, pp. 145–159.
Turner, A. (2007) ‘From axial to road-centre lines: a new representation for space syntax and a new model of route
choice for transport network analysis’, Environment and Planning B: Planning and Design, 34(3), pp. 539–555.
doi: 10.1068/b32067.
Turner, A. (2009) ‘Stitching Together the Fabric of Space and Society: An Investigation into the Linkage of the Local
to Regional Continuum’, in Daniel, K., Marcus, L., and Steen, J. (eds) Proceedings of the Seventh International
Space Syntax Symposium. Stockholm: KTH Royal Institute of Technology, pp. 1–12. Available at: http://eprints.
ucl.ac.uk/16184/ (Accessed: 24 June 2013).
Turner, A., Penn, A. and Hillier, B. (2005) ‘An algorithmic denition of the axial map’, Environment and Planning B:
Planning and Design, 32(3), pp. 425–444. doi: 10.1068/b31097.
Proceedings of the 11th Space Syntax Symposium
150.26
EMPLOYING VOLUNTEERED GEOGRAPHIC INFORMATION IN SPACE SYNTAX ANALYSIS
Varoudis, T., Law, S., Karimi, K., Hillier, B. and Penn, A. (2013) ‘Space Syntax Angular Betweenness Centrality
Revisited’, in Kim, Y. O., Park, H. T., and Seo, K. W. (eds) Proceedings of the Ninth International Space Syntax
Symposium. Seoul: Sejong University, pp. 1–16.
Zielstra, D. and Zipf, A. (2010) ‘A Comparative Study of Proprietary Geodata and Volunteered Geographic Information
for Germany’, in 13th AGILE International Conference on Geographic Information Science. Guimarães, Portugal,
pp. 1–15. doi: 10.1119/1.1736005.
