Cities 89 (2019) 14–25

Contents lists available at ScienceDirect

Cities
journal homepage: www.elsevier.com/locate/cities

A computational approach to ‘The Image of the City’ T

a,b,* a b,c
Gabriele Filomena , Judith A. Verstegen , Ed Manley
a
Institute for Geoinformatics, University of Münster, Heisenbergstraße 2, Münster 48149, Germany
b
The Bartlett Centre for Advanced Spatial Analysis, University College London, 90 Tottenham Court Rd, London W1T 4TJ, United Kingdom
c
The Alan Turing Institute, 96 Euston Rd, Kings Cross, London NW1 2DB, United Kingdom

ARTICLE INFO ABSTRACT

Keywords: In The Image of the City Lynch describes how individuals perceive and recall features in urban spaces. The most
Image of the City distinctive elements in the urban landscape - categorised in paths, nodes, edges, districts and landmarks - give
Cognitive maps shape to individuals' mental representation of the city. Lynch’s approach has stimulated research into spatial
Kevin Lynch cognition, urban design and artificial intelligence, and it still represents an essential pillar in the analysis of
Street network
urban dynamics. Nevertheless, an explicit link between The Image of the City and GIScience has not been com-
GIScience
pletely explored yet. In this paper, a computational approach to The Image of the City is proposed. Different
perspectives in spatial cognition and GIS research are integrated to obtain a complete Image of the City, in which
the most salient elements are shared by a large part of citizens. Nodes, paths and districts were identified through
network science techniques. Methods drawn from the information approach to The Image of the City are used to
detect landmarks, integrating the complexity of points of reference in their visual, structural and semantic
components, as conceptualised by Lynch and successive research. The methods were applied to the central area
of Boston and built using freely available spatial datasets. Results were compared to Lynch’s maps to evaluate the
methodology: besides a considerable discrepancy with regard to landmarks, a good correspondence for paths,
nodes, edges and districts was found.

1. Introduction knowledge organisation in landmark, route, and configurational

knowledge (Siegel & White, 1975). Successively, Lynch’s model has
In 1960, Kevin Lynch published The Image of the City (Lynch, 1960), been reformulated in behavioural geography (e.g. Sadalla, Burroughs, &
one of the most influential theories in spatial cognition and behavioural Staplin, 1980; Couclelis, Golledge, Gale, & Tobler, 1987; Tversky,
geography ever written (Portugali, 2011). Lynch dedicates his work to 1993). The Image of the City has furthermore inspired research in en-
understanding how people perceive and represent a city, and from what vironmental psychology (Devlin, 2001) and prompted the development
external urban artefacts the mental image of the city arises. Specifically, of contemporary cognitive mapping research (Tversky, 2003; Kitchin &
Lynch analyses two qualities of the built environment: legibility, ‘the Blades, 2002): cognitive maps are seen as auto-organising systems that
ease with which its parts can be recognised and can be organised into a mediate the human-environment interaction (Portugali, 1992).
coherent pattern’ and imageability, ‘that quality in an object which gives Inevitably, this framework has stimulated research in artificial in-
it a high probability of evoking a strong image in’ the observer’ (Lynch, telligence and robotics, where computational models such as TOUR
1960, p. 60). (Kuipers, 1978) and NAVIGATOR (Gopal, Klatzky, & Smith, 1989) have
Studying the city form of Boston, Los Angeles and Jersey City, the been developed to reproduce and comprehend human mental pro-
author finds out that individuals’ mental images can be overlapped to cesses. Nevertheless, there has not been a thorough attempt to integrate
form a community cognitive map. The resulting skeleton is formed by Lynch’s theory directly into GIScience, nor to support a quantitative
five types of elements: paths, edges, nodes, districts and landmarks. The formulation of The Image of the City relying on conventional spatial
mental image mediates in the interaction between humans and their datasets. ‘A generalisation of these (Lynch’s) ideas could advance geo-
environment: ‘The environment suggests distinctions and relations and graphic thinking in general and GIS software in particular’ (Frank,
the observer (..), selects and organizes and endows with meaning what Mark, & Raubal, 2013, p. 3).
it sees’ (Lynch, 1960, p. 6). Since the 1970s, the five elements have The aim of this work is to provide such a quantitative formulation of
constituted the core of spatial cognition theories, grounding the spatial The Image of the City to foster an inclusion of the concept of The Image of

*
Corresponding author.
E-mail address: [email protected] (G. Filomena).

https://doi.org/10.1016/j.cities.2019.01.006
Received 21 August 2018; Received in revised form 5 December 2018; Accepted 3 January 2019
Available online 21 January 2019
0264-2751/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
G. Filomena, et al. Cities 89 (2019) 14–25

the City in GIScience. In the first part of this paper, a set of works that Their theory reinvigorates the idea that city elements may be re-
have attempted to reformulate Lynch’s theory is reviewed. Following membered for symbolic and social meanings (Appleyard, 1969), but
this, an integration of such different reformulations is devised. In the also for their pragmatic functions. Moreover, it introduces a computa-
second part, a computational method, built upon conventional geos- tional framework to The Image of the City that takes into account the
patial data, is proposed for each Lynchian element in reference to the combination of these traits and their contribution to the legibility of
original definitions. The methods are applied to the central area of cities. The drawback of their approach is that it contemplates a measure
Boston and thereafter the results are discussed and subjectively com- of entropy, suitable for the city- or district-level but unfitting for
pared to Lynch’s qualitative maps. computing the individual scores of the elements.

2. Background 2.3. Automatic landmark extraction

A number of research communities have been involved in the Sorrows and Hirtle (1999) refine the notion of landmark. The au-
quantitative formulation of memorability and in extracting cognitively thors differentiate visual landmarks - objects used as spatial points of
salient urban features. These approaches can potentially contribute to reference for their visibility - from cognitive landmarks - relevant for
the development of a framework for extracting Lynch’s five elements. their uncommon and meaningful content - and structural landmarks -
recognisable for their advantageous and prominent position in the
2.1. Space Syntax space. This work has inspired a vein of research in GIS community in-
terested in automatic identification of landmarks for wayfinding design
The relationship between the external space and social phenomena and navigational support: Raubal and Winter (2002) advance a model
is the driving force of Space Syntax, a set of theories and techniques ‘for that measures the salience of buildings in relation to perceptual and
the representation, quantification, and interpretation of spatial config- cognitive properties. Winter (2003) ameliorates the model, re-
uration in buildings and settlements’ (Hillier, Hanson, & Graham, 1987, commending to consider 3d visibility as another property. Elias (2003)
p. 363). In this perspective, the street layout and the configuration of presents a similar approach based on machine learning algorithms that
space have a strong impact on the development of mental representa- inspect geometric, topological and semantic attributes of buildings to
tions (Kim & Penn, 2004). The association between street configuration establish landmark hierarchies. Furthermore, Winter, Tomko, Elias, and
and cognitive mapping is not unprecedented. What distinguishes Space Sester (2008) integrate the previous approaches to construct a hier-
Syntax research is the focus on topology rather than metric properties. archy of landmarks, emphasising the distinction between local-land-
Within this community, Dalton and Bafna (2003) attempt to re- marks and city-wide (global) landmarks.
define Lynch’s five elements through the constructs of axial lines In summary, while these works have moved landmark research
(Hillier & Hanson, 1984) and isovists (Benedikt, 1979). They suggest to forward, the other Lynchian elements are of little interest and rarely
study and detect first order (spatial and structural) elements, that give mentioned here.
shape to the mental representation, and second order (visual) elements,
that enrich the image. The elements could be captured employing axial 2.4. Contributions and gaps
lines and isovists sorted in order of significance.
Jiang (2013) theorises that the scaling law of artefacts (Zipf, 1949) Even though automatic landmark identification models have been
supports the identification of the primary elements in the city. He tested disseminating (Richter, 2013), presumably due to their potential ap-
his hypothesis on paths, transforming streets into axial lines and plicability in navigation systems design, none of the approaches have
ranking them by connectivity. Consistently with his assumptions, many offered a set of methods and tools to quantitatively derive the five
more less connected than well-connected and memorable streets exist, elements of The Image of the City.
regardless the city examined and the street morphology.
The Space Syntax approach has been questioned for excluding me- • Space Syntax, when reformulating The Image of the City, has mostly
tric information from the analysis, heights of buildings and land use considered visual aspects, neglecting important implications re-
properties from the analysis (Ratti, 2004). Moreover, although road- garding the genuine human-environment interaction and focused on
centre lines have been recently adopted (Turner, 2007), axial lines have paths identification.
obstructed the integration of space syntax techniques into GIS (Jiang & • The information approach, while being based on the usage of
Liu, 2009). Yet, the emphasis on configurational aspects is an illumi- geospatial dataset, returns a macro-level index of legibility.
nating argument in the transition towards a computational approach to • Edges have generally received a little attention, or been considered a
cognitive maps. particular type of landmarks and assessed for their structural
properties (e.g. Richter & Winter, 2014; Raubal & Winter, 2002);
2.2. The information approach districts have been translated in Voronoi partitions, whose cognitive
salience is disputable.
Haken and Portugali (2003) have advanced a framework to The • More importantly, in the literature discussed above, when an ap-
Image of the City that integrates the synergetics approach (Haken, 1983) plication of the methodologies is presented, the dataset usually re-
and contemporary trends in cognitive science with Lynch’s theory. The fers to small areas, it is created ad-hoc by the researchers or based
researchers make the link between the concepts of imageability and on questionnaires, which makes it hard to reproduce the study for
affordance (Gibson, 1979) explicit, on one side, and employ Shannon’s new areas.
information theory, on the other, for studying how the mental image is
formed from specific urban elements. In this information approach, 3. Methodology
legible cities are composed of informative and significant artefacts or
urban configurations. City nodes, paths or districts are ‘information In the following section, network science techniques are presented
carriers’ that shape the mental image. Here, the original measure of for the detection of nodes, paths and districts, from the street config-
information (Shannon & Weaver, 1949) - a form of entropy that uration. In addition, a comprehensive landmark detection method is
quantitatively measures the “unexpectedness” of an event - is adjusted proposed following Sorrows and Hirtle’s framework (Sorrows & Hirtle,
with an index that incorporates semantic information. This component 1999) and the models discussed above. These approaches were en-
represents the result of biological, cultural, social and pragmatic cate- riched performing a 3d visibility analysis and integrated with insights
gorisation processes. derived from the information approach; semantic and pragmatic

15
G. Filomena, et al. Cities 89 (2019) 14–25

properties were here considered for the first time in a large geo-dataset. major paths are those which minimise angular change in travels across
Lastly, a set of rules to extract edges is described. The scripts used to the city. Here, we did not take into account the road-type class (e.g.,
carry out the analysis are available on a GitHub repository (Filomena & major-road, secondary-road, etc.), assuming that the street segments
Manley, 2018). which form angular-continuing lines (e.g. Porta, Crucitti, & Latora,
2006a; Figueiredo & Amorim, 2005) belong to the same category or
3.1. Nodes have similar structural properties (Thomson, 2004).

Nodes are the strategic foci into which the observer can enter, and which 3.3. Districts
are the intensive foci to and from which he is travelling. They may be pri-
marily junctions, places of a break in transportation, a crossing or con- Districts are the relatively large city areas which the observer can men-
vergence of paths (Lynch, 1960, p. 47). tally go inside of, and which have some common character (Lynch, 1960, p.
Space Syntax scholars have shown that elements stored in people’s 66). The characteristics that determine districts are thematic continuities
cognitive map are related to centrality measures (Haq & Girotto, 2003; which may consist of an endless variety of components (Lynch, 1960, p.
Yun & Kim, 2007). In this framework, topological properties of degree, 67).
closeness and betweenness are usually computed employing a dual Space Syntax suggests that the topology of the street network is
graph representation, wherein street segments are represented by ver- associated with people’s perception of places and regions: Law (2017)
texes and junctions by links. However, Porta, Crucitti, and Latora illustrates a process for generating sub-graphs from the street topology
(2006b) argue that the primal representation approach - wherein street applying community detection techniques. The so formed Street-based
segments are represented by links and junctions by vertexes - is more Local Areas (SLA) (Turner, 2007) are regions whose internal homo-
effective in exploiting centrality indices to capture the skeleton of the geneity has social and functional foundations (Girvan & Newman,
urban structure and identify crucial intersections. 2002).
Centrality acts as a driving force in the development of the urban In this discussion, it is assumed that the different districts of a city
structure, and central locations are prone to become genuine urban can be identified analysing the road layout. The modularity optimisation
nodes (Newman & Kenworthy, 1999). Lynch’s nodes could be viewed as function (Blondel, Guillaume, Lambiotte, & Lefebvre, 2008) is adopted
places that are structurally made to be traversed, nodes (vertexes) with to extract SLAs. This algorithm optimises modularity (Girvan &
the highest betweenness centrality scores in the street network. In our Newman, 2002), an index that measures the goodness of a network
approach betweenness centrality was calculated in an undirected planar division. Modularity (Q) computes the difference between the edges
graph, as: within a community and the expected numbers of edges in a network
with the same structure but random connections. When the number of
njk (i) within-community edges is nothing more than random, the structure of
CiB =
j , k G, j k i
njk (1) the communities is poor and Q is equal to zero. On the contrary, high
values of Q represent strong division amongst well-structured commu-
where, in an undirected graph G, njk is the number of shortest paths nities. The implementation of the modularity optimisation technique
between the vertexes j and k, and njk(i) is the number of shortest paths follows these steps:
between the vertexes j and k that pass through i.
1. Every node i is considered a community.
3.2. Paths 2. For each node i, the algorithm evaluates the gain in modularity (Q)
that would be obtained by joining the node with each of the
Paths are the channels along which the observer customarily, occasion- neighbour communities j.
ally, or potentially moves. (..). People observe the city while moving through 3. If no possible gain is detected, the node i stays in its original com-
it, and along these paths the other environmental elements are arranged and munity, otherwise, it is placed in the community wherein the
related (Lynch, 1960, p. 47). modularity gain would be maximised.
Paths are the main lines of movement in the city; they guide peo- 4. The nodes in the same community form a new super vertex.
ple’s movement, supporting orientation. In low-legibility contexts, 5. The previous steps are repeated until the modularity cannot be
when a path is not characterised by vivid activities or peculiar prop- optimised any more (merging communities does not produce posi-
erties, perceptual continuity comes into play: people rely on this tive gain).
functional quality to successively travel across the city. The concept of
continuity recalls the idea that people tend to choose routes that The function was run in an undirected dual graph. Partitions were
minimise angular change rather than distance (Sadalla & Montello, extracted from a network where weights were based on the angles of
1989; Golledge, 1995; Hillier & Iida, 2005). In this sense, angular be- incidence between pairs of segments, as described above. Thereby,
tweenness is described as the best predictor of pedestrian and vehicular districts are the sub-graphs obtained from the street network by opti-
movement when only the street network is at disposal of the researcher mising modularity.
(Cooper, 2015). Betweenness centrality, as defined above (see Eq. (1)),
was computed for vertexes in a dual graph representation so generated: 3.4. Landmarks

• Street segments are converted to vertexes; Landmarks are point references considered to be external to the observer.
• When two street segments cross each other in the road network, a They are more easily identifiable, if they have a clear form; if they contrast
link connecting the corresponding vertex in the dual representation with their background; and if there is some prominence of spatial location
is created. (Lynch, 1960, pp. 78–79).
• The amplitude of the angle of incidence formed by two street seg- The present work follows Sorrows and Hirtle’s framework (Sorrows
ments is assigned to the corresponding link as weight. & Hirtle, 1999) in the distinction of three type of landmarks: visual,
structural and cognitive (pragmatic and cultural meanings). Indeed, for
Finally, the centrality values of the vertexes in such a network were Lynch, an edifice may become a landmark when it stands out from the
reassigned to the originating segments. Therefore, it is assumed that the background. Additionally, activities and historic references may

16
G. Filomena, et al. Cities 89 (2019) 14–25

contribute to reinforcing the legibility when visual attraction is in- itineraries and experiences, and recognise places (see Fig. 2).
sufficient. Raubal and Winter’s (Raubal & Winter, 2002) set of techni- Our analysis mainly relies on the street network and buildings
ques were enriched to allow for topological relations, visibility, se- footprints. While the datasets are Boston-specific, the sources employed
mantic and pragmatic aspects in landmarks identification. Our analysis are generic (and often available as open data) so to allow application of
was performed on a dataset containing exclusively buildings; other this approach for other case study areas. The sources are:
elements such as trees, benches or bridges were not considered.
Visual properties includes height, façade area and visibility. The • The road network (Massachusetts Office of Geographic and
maximum height of a building was used for computing a 3d visibility Environmental Information, 2014), clipped with two buffers of 4000
index: for each edifice, the length of the longest unobstructed sight-line and 8000 m around Boston City Hall. The smaller network was used
was kept and used as a coarse value of visibility. For what concerns to identify nodes and paths, the vaster to detect regions.
structural and topological properties, advance visibility - a 2d area of • Building footprints (Massachusetts Office of Geographic and
visibility, without obstructions, around a building -, minimum distance Environmental Information, 2002); this dataset includes a simplified
from the road, the number of adjacent buildings (at least partially representation of buildings footprints, along with their maximum
contained in a buffer of x metres) and the area of the polygon were height.
computed. The cultural meaning of a building was obtained counting • Buildings main land use (OpenStreetMap contributors, 2017) and
the number of listed historic elements located within its boundaries. Boston Historical landmarks register (City of Boston, 2017). The
Finally, pragmatic significance was calculated following a simplifica- land-use classes used to compute the pragmatic score were: attrac-
tion of the information approach, as an index of unexpectedness: tion, commercial, cultural, eating & drinking, education, emergency
service, entertainment, hospitality, industrial, library, manufacturing,
Nb
Psb = 1 medical care, military, place of worship, public, residential, sport,
N (2)
transport, university.
where, in a buffer of x metres around the building b, Nb is the fre- • Railways (Massachusetts Office of Geographic and Environmental
quency of the land use class of b and N is the number of buildings. The Information, 2015) and water (Massachusetts Office of Geographic
scores of the indexes were scaled and combined in the relative com- and Environmental Information, 2017), to detect edges.
ponent and, subsequently, in the overall score (see Table 1 for details).
We subjectively defined weights for each landmark component, as
3.5. Edges specified in Table 1, coherently with Lynch’s observations. Yet, the
weights definition may vary with the urban structure, the social context
Edges are linear elements not considered as paths: they are usually, and so forth, and can be manipulated correspondingly for other case
boundaries between two kinds of areas. They act as lateral references. Those studies. Finally, to obtain the overall Image of the City, the range of
edges seem strongest which are not only visually prominent, but also con- original values was rescaled in the range from 0 to 1, on the bases of the
tinuous in form and impenetrable to cross movement (Lynch, 1960, p. 62). maximum and minimum features across each element; the ranges used
Edges are authentic organising features whose primary trait is linear by Lynch were applied to the scaled scores of nodes, paths and land-
continuity. Nevertheless, edges could be permeable and crossable, they marks, to colour and rank them. Due to the nature of the methods,
can coincide and align with paths. In the current analysis, the following districts and edges were not ranked.
linear elements, with a predefined minimum length, were extracted as
edges: 4. Results

• Sections of railway structures as bypasses or other visible structures. 4.1. Nodes, paths and districts

• Sections of large roads (e.g. dual-carriageways roads).

• Sections of motorways. We found some crucial nodes (Fig. 3) in the city as Charles Street

• River banks or generic waterfronts (lakes, sea-coast). Rotary and the junction between State Street and Congress Street. The
complex whole of ramps and interchanges nearby Charles River Road
and Martha Road confers high betweenness centrality values to other
3.6. The case study intersections in the area: this cluster of nodes is represented by Lynch as
one node at the south-east extremity of Charles River Road (see Fig. 2).
The methods delineated above were applied to the city centre of Likewise, the Sumner and Callahan tunnels, essential links to and from
Boston, MA (USA), on the area studied by Lynch (Fig. 1). The results of East Boston, give emphasis to the nodes at their entrances and the
the analysis are presented, discussed and compared to the map depicted nearby junction (Congress and North Street).
by Lynch, who asked 30 residents and workers to describe customary Like Lynch, we detect a lack of main nodes in the western area of the
city. Nonetheless, the output of our analysis does not point Worcester
Square and Union Park as crucial nodes, nor the intersections of
Table 1
Landmark extraction: indexes and weights. Huntington Avenue and Columbia Avenue, respectively with St James
Avenue and Stuart Street. Besides, Copley Square, Scollay Square and
Component Index Index weight Component weight
South Station do not stand out as they do in Lynch map (see Fig. 2). While
3d Visibility 0.50 most of this nodes were mentioned by just a few interviewees (12–25 %
Visual Façade area 0.30 0.50 cluster), South Station (over 75%) and Copley Square (50–75 %) may
Height 0.20 gain their importance in citizens’ mental maps for their role in the public
transport network, which is not taken into account in our analysis.
Area 0.30
2d advance visibility 0.30
Lynch describes the system of paths in Boston as confused but,
Structural 0.30
Neighbours (150-mt buffer) 0.20 notwithstanding, functional: major movement lines shape the overall
Road distance 0.20 Image of the City, converging in the city centre. Here, the motorway
may be seen as a constraining and limiting element, endowed with
Semantic Historical importance 1.0 0.10
organising properties rather than being meaningful for movement.
Pragmatic Land use (200-mt buffer) 1.0 0.10 Angular edge betweenness captured the main paths detected by
Lynch as Beacon Street, Boylston Street, Cambridge Street, Charles

17
G. Filomena, et al. Cities 89 (2019) 14–25

So
rro
wD
riv
e

Fig. 1. The city centre of Boston and the surrounding areas.

Source: Mapbox, OpenStreetMap.

Street, Tremont Street and Massachusetts Avenue (Fig. 4). Nevertheless, due to peculiar buildings, agglomerated activities and the presence of
Lynch also depicts as major paths Atlantic Avenue and the narrow ethnic communities, cannot be completely captured by network techni-
Washington Street. While the first receives its importance from the sea ques. Since the motorways and the system of interchanges caused some
bank and the harbour, Washington Street owes its legibility to the high distortions, a network composed only of local roads was also used in this
edifices along it. The fact that architectural properties and environ- case; the results are displayed in the Appendix (Fig. A1)
mental evaluations were not taken into account in our paths identifi-
cation approach may explain this dissimilarity. In addition, our method 4.2. Landmarks and edges
did not identify Storrow Drive, a road also perceived as a barrier by
Lynch’s interviewees. According to Lynch, in Boston, edifices are perceived as singular
Districts in Boston are vivid and orientating entities: even though entities when the city centre is admired from the banks of Charles River
their structure may be confusing and unclear from a purely structural and only a few of them have enough vividness to be identified as singular
point of view, their thematic identity is strong (Lynch, 1960). Because point of references. In particular, the prominence of State House, Old
of their connection to personal experiences and activities, many Lynch’s John Hancock Building (nowadays Berkeley Building) and Custom House
interviewees indicated districts in Boston as the main elements in the on the rest of city was mostly noticed from external points of view: their
mental representation of the city. These areas have central cores and relation with the road was not as strong as their visibility and people
undefined boundaries, sort of ‘thematic gradient’, that vanish gradually preferred to take advantage of local landmarks (Lynch, 1960).
rather than precisely. In our landmark results (Figs. 6 and A2 for sub-component scores),
In our results, Back Bay, which morphologically exhibits an un- Prudential Center, John Hancock Tower, Berkeley Building, Copley
common regularity compared to the rest of the city, the core of North Tower, Massachusetts General Hospital, South Station and the T-Garden
End-Docks area and the Financial District clearly stand out from the rest Arena, came out as main landmarks, along with a large number of local
of the city centre (Fig. 5). At the same time, the triangle between Back landmarks. However, besides the Massachusetts General Hospital and
Bay and South End, portrayed in The Image of the City as an empty area the Berkeley Building, Lynch also indicates some other major land-
devoid of any character, is extracted as a single entity (coloured in marks - as State House, Christian Science Church, the Public Library,
fuchsia) by the algorithm. These portions almost coincide with Lynch’s the Trinity Church and Custom House - that did not particularly dis-
outcomes and considerations. On the contrary, Beacon Hill and West End tinguish themselves in our analysis.
are here merged in a macro-region, while some micro-districts as Thea- Whereas major lines of movement, street configuration and nodes
tres, China Town or the Textile Leather area, are assimilated by the Fi- may have slightly changed since Lynch’s original study, the city centre
nancial/Shopping district in our results. Their vivid connotation, possibly has undergone a more consistent evolution in terms of vertical

18
G. Filomena, et al. Cities 89 (2019) 14–25

Fig. 2. The Boston community map as emerged by verbal interviews.

Source: Lynch, Kevin, The Image of the City, Figure 35 and corresponding legend, pp. 145–146, 1960. Copyright Massachusetts Institute of Technology, by per-
mission of The MIT Press.

Fig. 3. Nodes: street junctions coloured by betweenness centrality. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)

19
G. Filomena, et al. Cities 89 (2019) 14–25

Max

Min

Fig. 4. Paths: street segments coloured by angular edge betweenness. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)

Fig. 5. Districts: Street-based Local Area (SLA) in Boston city centre. Each colour represents streets belonging to a single district; semi-transparent colours indicate
streets outside the studied area. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

20
G. Filomena, et al. Cities 89 (2019) 14–25

Max

Min

Fig. 6. Landmarks: buildings coloured by the overall score. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)

development. The large number of skyscrapers erected throughout the motorways, besides reinforcing the peninsula profile, represent inter-
last 60 years along with the general evolution of the city and a possible ruptions, separate different areas and obstruct movement. The Central
redistribution of the activities, partly explain the discrepancies between Artery is described by Lynch’s interviewees as a fragmentary and ab-
Lynch’s and our landmark scores. Interestingly, such differences may stract edge: even though sometimes it is not visible, people are aware of
disclose the dynamic nature of the mental image of the city, a re- its presence and the impossibility of crossing it. North End is a shiny
presentation that varies and readapts, embracing new elements, example in this sense. The Central Artery completely separates this area
meanings and possibilities. from the rest of the city centre.
In Boston, edges seem to play two roles. On one hand, the river and The edges identified in our analysis (Fig. 7) correspond approxi-
the harbour define the shape of the city centre; on the other hand, the mately with those discussed in The Image of the City. The waterfront is

Railways
Highways
Waterfront

Fig. 7. Primary edges in Boston. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

21
G. Filomena, et al. Cities 89 (2019) 14–25

Fig. 8. The computational Image of the City of Boston.

automatically depicted as a continuous and uninterrupted edge that changes occurred in the city in the last 60 years, there were corre-
draws the main contour of the cognitive representation of the city. spondences between the other elements extracted computationally and
However, Lynch illustrates how citizens rarely join Charles River bank those reported by Lynch: firstly, centrality measures consented to detect
with the harbour. The water between West End and North End is hidden main nodes from the street network, as well as major paths; secondly,
by several buildings and railway structures, which make the continuity the modularity optimisation algorithm, a community-detection tech-
of the waterfront difficult to conceptualise and mentally represent. nique, subdivided the area studied in portions alike to Lynch’s main
From our analysis, the interplay between the river and Storrow Drive districts; lastly, the criteria advanced to pull out edges produced sa-
emerges as well: the latter reinforces the meaning of the first but, tisfying results, against the ones presented in The Image of the City.
nevertheless, causes a sense of distance between the water and the Having said that, the complexity of human cognition and perception
districts of Back Bay and Beacon Hill. Conversely, the edge represented cannot be fully captured with a computational approach to mental re-
by the harbour is more fluid: the presence of amenities and flourishing presentations. Interviews and sketch-maps are more comprehensive
activities make this area a vivid and liveable edge. tools when it comes to study humans’ representations, their experiences
Finally, the computational Image of the City of Boston, displayed in and their decisions within the environment. The fact that the waterfront
Fig. 8, was obtained combining all the elements in a single, overall map of the peninsula was identified as an edge, not accounting for the ab-
in the style of the hand-drawn maps created by Lynch. sence of perceptual continuity between West End and North End, shows
the elusiveness of human cognition. Yet, there are various routes along
5. Discussion and conclusion which our work might be extended. The public transport network
should be considered for determining crucial nodes, beyond the street
The aim of this work was to provide a quantitative formulation of junctions, by means of multi-layered or multiplex network. In this
Lynch’s Image of the City, easily incorporable in GIS environments, that sense, Tomko and Winter (2013) advance a formal approach to urban
may favour a more explicit inclusion of The Image of the City in form representation, wherein they suggest to extract The Image of the
GIScience. In The Image of the City, Lynch introduces and describes five City taking into account the mean transport used by the observer.
elements - nodes, paths, districts, landmarks and edges - that give shape Moreover, the definition of districts may be adjusted through in-
to the mental representation of the city. A complete computational corporation of demographic and social dimensions (for example, Gao,
approach to The Image of the City was presented here and tested on a Janowicz, & Couclelis, 2017; Gao, Janowicz, Montello et al., 2017 have
large and freely available urban-dataset, integrating a range of methods extracted urban functional regions employing Point-Of-Interests (POIs)
derived from previous research. The mental image of Boston was drawn check-ins and social-media data). Similarly, the definition of edges may
ranking artefacts on the basis of network and geospatial measures. We require a more precise formalisation and the incorporation of elements
explicitly took into consideration a semantic component in landmark of structural change in the urban morphology.
extraction and tested a series of criteria to pick out edges. The results This framework may also undergo a more precise validation. To
were visually compared with the maps of Boston presented in The Image further generalise the validity of the approach here presented, the
of the City, outputs of Lynch’s qualitative analysis. While it was difficult methodology should be tested and evaluated with other case-studies,
to find a common ground concerning landmarks, possibly due to the allowing for differences in urban planning approaches. Origin-

22
G. Filomena, et al. Cities 89 (2019) 14–25

destination matrices regarding pedestrian and traffic flows, or POI areas (activities, movement and orientation).
check-in datasets, might give indications on paths, nodes, or re- The work has shown that the development of a computational form
gionalisation (Manley, 2014), or indicate how to calibrate the weights of The Image of the City is feasible. The methodology devised here can be
in the landmark extraction. applied to other cities and urban contexts very easily, manipulating just
The model developed during this paper can be manipulated in any the input data and a few parameters. This approach makes it possible to
GIS package, enabling formal spatial analysis and modelling of the re- reveal images of cities, investigated so far with qualitative and time-
lationship between observed spatial behaviour and urban form. consuming procedures. We argue that this tool may support spatial
Examples might include the development of new simulation models of planning decisions in urban design, providing important insights as
pedestrian or vehicular movement (e.g. Manley, Orr, & Cheng, 2015), concerns city livability, quality of life (Llinares, Page, & Llinares, 2013),
spatial and accessibility planning, and identification of low-legibility the adequate mix of land-uses, the ease of navigation and orientation.

Appendix

Fig. A1. Districts: street-based Local Area (SLA) in Boston city centre extracted from a street-network without motorways. Each colour represents streets belonging to
a single district; semi-transparent colours indicate streets outside the studied area. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)

23
G. Filomena, et al. Cities 89 (2019) 14–25

Fig. A2. Landmarks: buildings coloured by component scores. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)

References International Journal of Geographical Information Science, 31(6), 1245–1271. https://

doi.org/10.1080/13658816.2016.1273357.
Gibson, J. J. (1979). The ecological approach to visual perception. Boston, MA: Houghton
Appleyard, D. (1969). Why buildings are known. Environment and Behavior, 1(2), Mifflin.
131—15 6. Girvan, M., & Newman, M. E. J. (2002). Community structure in social and biological
Benedikt, M. L. (1979). To take hold of space: Isovists and isovist fields. Environment and networks. Proceedings of the National Academy of Sciences of the United States of
Planning B: Planning and Design, 6, 47–65. America, 99(12), 7821–7826. https://doi.org/10.1073/pnas.122653799.
Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of Golledge, R. G. (1995). Path selection and route preference in human navigation: A
communities in large networks. Journal of Statistical Mechanics: Theory and progress report. In A. U. Frank, & W. Kuhn (Vol. Eds.), Spatial Information Theory A
Experiment, 2008(10), 1 —12. https://doi.org/10.1088/1742-5468/2008/10/ Theoretical Basis for GIS. COSIT 1997. Lecture Notes in Computer Science. Vol.1329 (pp.
P10008. 207–222). Berlin: Springer. https://doi.org/10.1007/3-540-60392-1_14.
City of Boston. (2017). Boston Landmark Commission. Retrieved from http:// Gopal, S., Klatzky, R. L., & Smith, T. R. (1989). Navigator: A psychologically based model
bostonopendata-boston.opendata.arcgis.com/datasets/ of environmental learning through navigation. Journal of Environmental Psychology,
7a7aca614ad740e99b060e0ee787a228_3 (Accessed 2017-01-11) 9(4), 309–331. https://doi.org/10.1016/S0272-4944(89)80012-X.
Cooper, C. H. (2015). Spatial localization of closeness and betweenness measures: A self- Haken, H. (1983). Synergetics: An introduction. Nonequilibrium phase transitions and self-
contradictory but useful form of network analysis. International Journal of organization in physics, chemistry, and biology. Berlin: Springer.
Geographical Information Science, 29(8), 1293–1309. https://doi.org/10.1080/ Haken, H., & Portugali, J. (2003). The face of the city is its information. Journal of
13658816.2015.1018834. Environmental Psychology, 23(4), 385–408. https://doi.org/10.1016/S0272-4944(03)
Couclelis, H., Golledge, R. G., Gale, N., & Tobler, W. (1987). Exploring the anchor-point 00003-3.
hypothesis of spatial cognition. Journal of Environmental Psychology, 7(2), 99–122. Haq, S., & Girotto, S. (2003). Ability and intelligibility: Wayfinding and environmental
https://doi.org/10.1016/S0272-4944(87)80020-8. cognition in the designed environment. In J. Hanson (Ed.). Proceedings, 4th interna-
Dalton, R. C., & Bafna, S. (2003). The syntactical image of the city: A reciprocal definition tional space syntax symposium (pp. 68.1–68.20). London.
of spatial elements and spatial syntaxes. Proceedings, 4th International Space Syntax Hillier, B., & Hanson, J. (1984). The social logic of space. Cambridge, UK: Cambridge
Symposium (pp. 59.1–59.22). London. University Press.
Devlin, A. S. (2001). Mind and maze: Spatial cognition and environmental behavior. Hillier, B., Hanson, J., & Graham, H. (1987). Ideas are in things: An application of the
Westport, CT: Praeger Publishers/Greenwood Publishing Group. space syntax method to discovering house genotypes. Environment and Planning B:
Elias, B. (2003). Extracting landmarks with data mining methods. In W. Kuhn, M. F. Planning and Design, 14(4), 363–385. https://doi.org/10.1068/b140363.
Worboys, & S. Timpf (Vol. Eds.), Spatial Information Theory. Foundations of Geographic Hillier, B., & Iida, S. (2005). Network effects and psychological effects: A theory of urban
Information Science. COSIT 2003. Lecture Notes in Computer Science. Vol.2825 (pp. movement. In A. G. Cohn, & D. M. Mark (Vol. Eds.), Spatial Information Theory. COSIT
375–389). Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-39923- 2005. Lecture Notes in Computer Science. Vol.3693 (pp. 553–564). Berlin, Heidelberg:
0_25. Springer. https://doi.org/10.1007/11556114_30.
Figueiredo, L., & Amorim, L. (2005). Continuity lines in the axial system. 5th International Jiang, B. (2013). The Image of the City out of the underlying scaling of city artifacts or
Space Syntax symposium (pp. 161–174). Delft. locations. Annals of the Association of American Geographers, 103(6), 1552–1566.
Filomena, G., & Manley, E. (2018). Computational-Image-of-the-City. Retrieved from https://doi.org/10.1080/00045608.2013.779503.
https://zenodo.org/record/2541563https://doi.org/10.5281/zenodo.2541563. Jiang, B., & Liu, C. (2009). Street-based topological representations and analyses for
Frank, A. U., Mark, D. M., & Raubal, M. (2013). Researching cognitive and linguistic predicting traffic flow in GIS. International Journal of Geographical Information Science,
aspects of geographic space: Las Navas then and now. In M. Raubal, D. M. Mark, & A. 23(9), 1119–1137. https://doi.org/10.1080/13658810701690448.
U. Frank (Eds.). Cognitive and Linguistic Aspects of Geographic Space (pp. 1–22). Berlin: Kim, Y. O., & Penn, A. (2004). Linking the spatial syntax of cognitive maps to the spatial
Springer. https://doi.org/10.1007/978-3-642-34359-9_1. syntax of the environment. Environment & Behavior, 36(4), 483–504. https://doi.org/
Gao, S., Janowicz, K., & Couclelis, H. (2017). Extracting urban functional regions from 10.1177/0013916503261384.
points of interest and human activities on location-based social networks. Kitchin, R. M., & Blades, M. (2002). The cognition of geographic space. London: I.B. Tauris.
Transactions in GIS, 21(3), 446–467. https://doi.org/10.1111/tgis.12289. Kuipers, B. J. (1978). Modeling spatial knowledge. Cognitive Science, 2(2), 129–153.
Gao, S., Janowicz, K., Montello, D. R., Hu, Y., Yang, J.-A., McKenzie, G., ... Yan, B. (2017). https://doi.org/10.1207/s15516709cog0202_3.
A data-synthesis-driven method for detecting and extracting vague cognitive regions. Law, S. (2017). Defining street-based local area and measuring its effect on house price

24
G. Filomena, et al. Cities 89 (2019) 14–25

using a hedonic price approach: The case study of Metropolitan London. Cities, 60, Raubal, D. M. Mark, & A. U. Frank (Eds.). Cognitive and Linguistic Aspects of Geographic
166–179. https://doi.org/10.1016/j.cities.2016.08.008. Space. Lecture Notes in Geoinformation and Cartography (pp. 83–97). Berlin,
Llinares, C., Page, A., & Llinares, J. (2013). An approach to defining strategies for im- Heidelberg: Springer. https://doi.org/10.1007/978-3-642-34359-9_5.
proving city perception. Case study of Valencia, Spain. Cities, 35, 78–88. https://doi. Richter, K.-F., & Winter, S. (2014). Conceptual aspects: How landmarks can be described
org/10.1016/j.cities.2013.06.009. in data models. Landmarks (pp. 109–135). Cham: Springer International Publishing.
Lynch, K. (1960). The image of the city. Cambridge, MA: MIT Press. https://doi.org/10.1007/978-3-319-05732-3_4.
Manley, E. J. (2014). Identifying functional urban regions within traffic flow. Regional Sadalla, E. K., Burroughs, J. W., & Staplin, L. J. (1980). Reference points in spatial cog-
Studies, Regional Science, 1(1), 40–42. https://doi.org/10.1080/21681376.2014. nition. Journal of Experimental Psychology: Human Learning and Memory, 6(5),
891649. 516–528. https://doi.org/10.1037/0278-7393.6.5.516.
Manley, E. J., Orr, S., & Cheng, T. (2015). A heuristic model of bounded route choice in Sadalla, E. K., & Montello, D. R. (1989). Remembering changes in direction. Environment
urban areas. Transportation Research Part C: Emerging Technologies, 56, 195 —209. and Behavior, 21(3), 346–363. https://doi.org/10.1177/0013916589213006.
https://doi.org/10.1016/j.trc.2015.03.020. Shannon, C. E., & Weaver, W. (1949). The mathematical theory of communication. Urbana,
Massachusetts Office of Geographic and Environmental Information. (2002). Building IL: University of Illinois Press.
footprints and heights from LIDAR. Retrieved from https://earthworks.stanford.edu/ Siegel, A. W., & White, S. H. (1975). The development of spatial representations of large-
catalog/mit-us-ma-bostonmet-g46bldght-2002 (Accessed 2017-11-01) scale environments. Advances in Child Development and Behavior, 10, 9–55. https://
Massachusetts Office of Geographic and Environmental Information. (2014). doi.org/10.1016/S0065-2407(08)60007-5.
Massachusetts Department of Transportation (MassDOT) Roads. Retrieved from Sorrows, M. E., & Hirtle, S. C. (1999). The nature of landmarks for real and electronic
http://www.mass.gov/anf/research-and-tech/it-serv-and-support/application-serv/ spaces. In C. Freksa, & D. M. Mark (Vol. Eds.), Spatial information theory. Cognitive and
office-of-geographic-information-massgis/datalayers/eotroads.html (Accessed 2017- Computational Foundations of Geographic Information Science. COSIT 1999. Lecture
11-01) Notes in Computer Science. Vol.1661 (pp. 37–50). Berlin, Heidelberg: Springer. https://
Massachusetts Office of Geographic and Environmental Information. (2015). Trains. doi.org/10.1007/3-540-48384-5_3.
Retrieved from http://www.mass.gov/anf/research-and-tech/it-serv-and-support/ Thomson, R. C. (2004). Bending the axial line: Smoothly continuous road centre-line
application-serv/office-of-geographic-information-massgis/datalayers/trains.html segments as a basis for road network analysis. Proceedings of the 4th International
(Accessed 2017-11-01) Space Syntax Symposium. London.
Massachusetts Office of Geographic and Environmental Information. (2017). MassDEP Tomko, M., & Winter, S. (2013). Describing the functional spatial structure of urban
Hydrography (1:25,000). Retrieved from http://www.mass.gov/anf/research-and- environments. Computers, Environment and Urban Systems, 41, 177–187. https://doi.
tech/it-serv-and-support/application-serv/office-of-geographic-information-massgis/ org/10.1016/j.compenvurbsys.2013.05.002.
datalayers/hd.html (Accessed 2017-01-11) Turner, A. (2007). From axial to road-centre lines: A new representation for Space Syntax
Newman, P., & Kenworthy, J. R. (1999). Sustainability and cities: Overcoming automobile and a new model of route choice for transport network analysis. Environment and
dependence. Washington, DC: Island Press. Planning B: Planning and Design, 34(3), 539–555. https://doi.org/10.1068/b32067.
OpenStreetMap contributors. (2017). Planet dump retrieved from https://planet.osm.org. Tversky, B. (1993). Cognitive maps, cognitive collages and spatial mental model. In A.
Retrieved from https://www.openstreetmap.org (Accessed 2017-11-01) Frank, & I. Campari (Vol. Eds.), Spatial Information Theory A Theoretical Basis for GIS.
Porta, S., Crucitti, P., & Latora, V. (2006a). The network analysis of urban streets: A dual COSIT 1993. Lecture Notes in Computer Science. Vol.716 (pp. 14–24). Berlin,
approach. Physica A, 369(2), 853–866. https://doi.org/10.1016/j.physa.2005.12. Heidelberg: Springer. https://doi.org/10.1007/3-540-57207-4_2.
063. Tversky, B. (2003). Structures of mental spaces: How people think about space.
Porta, S., Crucitti, P., & Latora, V. (2006b). The network analysis of urban streets: A Environment and Behavior, 35(1), 66–80. https://doi.org/10.1177/
primal approach. Environment and Planning B: Planning and Design, 33(5), 705–725. 0013916502238865.
https://doi.org/10.1068/b32045. Winter, S. (2003). Route adaptive selection of salient features. In W. Kuhn, M. Worboys, &
Portugali, J. (1992). Geography, environment and cognition: An introduction. Geoforum, S. Timpf (Vol. Eds.), Spatial Information Theory. Foundations of Geographic Information
23(2), 107–109. https://doi.org/10.1016/0016-7185(92)90009-S. Science. COSIT 2003. Lecture Notes in Computer Science. Vol.2825 (pp. 349–361).
Portugali, J. (2011). Complexity, Cognition and the City. Berlin, Heidelberg: Springer- Berlin, Heidelberg: Springer. https://doi.org/10.1007/978-3-540-39923-0_23.
Verlag. Winter, S., Tomko, M., Elias, B., & Sester, M. (2008). Landmark hierarchies in context.
Ratti, C. (2004). Space Syntax: Some inconsistencies. Environment and Planning B: Planning Environment and Planning B: Planning and Design, 35(3), 381–398. https://doi.org/10.
and Design, 31(4), 487–499. https://doi.org/10.1068/b3019. 1068/b33106.
Raubal, M., & Winter, S. (2002). Enriching wayfinding instructions with local landmarks. Yun, Y. W., & Kim, Y. O. (2007). The effect of depth and distance in spatial cognition.
In M. J. Egenhofer, & D. M. Mark (Vol. Eds.), Geographic Information Science. GIScience Proceedings, 6th International Space Syntax Symposium (pp. (49) 1—14). Istanbul.
2002. Lecture Notes in Computer Science. Vol.2478 (pp. 243–259). Berlin: Springer. Zipf, G. K. (1949). Human behaviour and the principles of least effort. Cambridge, MA:
https://doi.org/10.1007/3-540-45799-2_17. Addison-Wesley.
Richter, K.-F. (2013). Prospects and challenges of landmarks in navigation services. In M.

25