The morphology of the Arrival City - A global categorization based on literature Surveys and remotely sensed data

Abstract

When we think about living environments of the urban poor, slums might be the most immediate association. These slums evoke a more or less stereotype impression of built environments: complex, high dense alignments of small makeshift or run-down shelters. However, this perceived characteristic morphology is neither globally homogeneous nor is this perception covering morphologic appearances of urban poverty in a comprehensive way. This research provides an empirical baseline study of existing morphologies, their similarities and differences across the globe. To do so, we conceptually approach urban poverty as places which provide relatively cheap living spaces serving as possible access to the city, to its society and to its functions-so called Arrival Cities. Based on a systematic literature survey we select a sample of 44 Arrival Cities across the globe. Using very high resolution optical satellite data in combination with street view images and field work we derive level of detail-1 3D-building models for all study areas. We measure the spatial structure of these settlements by the spatial pattern (by three features-building density, building orientation and heterogeneity of the pattern) and the morphology of individual buildings (by two features-building size and height). We develop a morphologic settlement type index based on all five features allowing categorization of Arrival Cities. We find a large mor-phologic variety for built environments of the urban poor, from slum and slum-like structures to formal and planned structures. This variability is found on all continents, within countries and even within a single city. At the same time detected categories (such as slums) are found to have very similar physical features across the globe.

Full text

Contents lists available at ScienceDirect
Applied Geography
journal homepage: www.elsevier.com/locate/apgeog
The morphology of the Arrival City - A global categorization based on
literature surveys and remotely sensed data
H. Taubenböck
∗
, N.J. Kraff, M. Wurm
German Aerospace Center (DLR), German Remote Sensing Data Center (DFD), Oberpfaffenhofen, Germany
ARTICLE INFO
Keywords:
Slums
Informal settlements
Urban poverty
Building morphologies
Urban pattern
Remote sensing
ABSTRACT
When we think about living environments of the urban poor, slums might be the most immediate association.
These slums evoke a more or less stereotype impression of built environments: complex, high dense alignments
of small makeshift or run-down shelters. However, this perceived characteristic morphology is neither globally
homogeneous nor is this perception covering morphologic appearances of urban poverty in a comprehensive
way. This research provides an empirical baseline study of existing morphologies, their similarities and differ-
ences across the globe. To do so, we conceptually approach urban poverty as places which provide relatively
cheap living spaces serving as possible access to the city, to its society and to its functions –so called Arrival
Cities. Based on a systematic literature survey we select a sample of 44 Arrival Cities across the globe. Using very
high resolution optical satellite data in combination with street view images and field work we derive level of
detail-1 3D-building models for all study areas. We measure the spatial structure of these settlements by the
spatial pattern (by three features –building density, building orientation and heterogeneity of the pattern) and
the morphology of individual buildings (by two features –building size and height). We develop a morphologic
settlement type index based on all five features allowing categorization of Arrival Cities. We find a large mor-
phologic variety for built environments of the urban poor, from slum and slum-like structures to formal and
planned structures. This variability is found on all continents, within countries and even within a single city. At
the same time detected categories (such as slums) are found to have very similar physical features across the
globe.
1. Introduction
“We did not see a mountain full of houses, but rather a house the size of
a mountain”(UTT, 2016). This statement captures the overwhelming
impression one gets of informal land occupation capitalizing every inch
of urban space in cities across the globe. Organic, amorphous, complex,
and dense seas of makeshift shelters have significantly different phy-
sical appearances than formal, planned parts in cities (e.g. Fig. 1). With
it, the built environment can be an expression of inequality in cities,
and socio-economic disparities even become visible from space (e.g.
Davis, 2007; Sliuzas, Mboup, & de Sherbinin, 2008). While a first su-
perficial observation may suggest forms of living at the lower end of
urban societies feature great similarities in terms of their physical ap-
pearance, (informal) processes such as illegal land occupation do not
always shape such distinct and demarcating building morphologies and
patterns for this social group (Saunders, 2010; Vaz & Berenstein, 2004).
Settlements of the urban poor are by no means a homogeneous
physical phenomenon (e.g. Schneider-Sliwa & Bhatt, 2008; Taubenböck
& Kraff, 2015). Nevertheless, most studies describing the physical ap-
pearance of such areas are of qualitative nature observing e.g. high
building densities or organic patterns as characteristic (e.g. Davis,
2007; Glaeser, 2010); however, relatively little systematic quantitative,
spatial research exists about their explicit physical appearance
(Hofmann, 2001; Kuffer, Pfeffer & Sliuzas, 2016), not to mention a
systematic inventory of morphologic types across the globe.
In this paper, we aim at reducing these knowledge gaps about set-
tlements of the urban poor by an empirical baseline study taking stock
of physical building types and determining structural patterns across
the globe. Avoiding terminological imprecision and related conceptual
restrictions of terms such as ‘slum’or ‘informal settlement’, we base this
study on the term ‘Arrival City’(introduced by Saunders, 2010). The
term conceptually integrates all places which provide comparably
cheap living spaces serving as possible access to the city, to its society
and to its functions for rural-urban migrants as well as for the existing
urban poor; this conceptual umbrella allows a broader perspective on
the specific places and their built morphologies.
https://doi.org/10.1016/j.apgeog.2018.02.002
Received 19 September 2017; Received in revised form 26 January 2018; Accepted 3 February 2018
∗
Corresponding author.
E-mail address: hannes.taubenboeck@dlr.de (H. Taubenböck).
Applied Geography 92 (2018) 150–167
0143-6228/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).
T

Arrival Cities show a large variety of built forms. Earth observation
(EO) data are the crucial data source to consistently capture built en-
vironments. However, due to conceptual imprecision of the target class,
due to challenges for image classifications algorithms in these complex
environments and due to unavailability or high costs of appropriate EO-
data, spatial data on the level of individual buildings (level of detail-1
(LoD-1)) for the neglected parts of cities are mostly inconsistent, gen-
eralized or simply inexistent. The purposeful development of
Fig. 1. The morphologic appearances of informal versus formal settlements for the example of Kibera, Nairobi; Top: Photography of Kibera:
©
Johnny Miller/Thomson Reuters
Foundation; Bottom: High resolution optical satellite imagery
©
Google Earth.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
151

classification algorithms for mapping living environments of the urban
poor from EO-data demands for empirical knowledge on the morphol-
ogies of the respective target class.
In this paper, we find 44 Arrival Cities by an extensive literature
survey. They function as representatives for built environments of
urban poor across the globe. For these samples, we produce three-di-
mensional (3D) building models in LoD-1 using a combination of high
resolution satellite data as well as auxiliary data sources (such as geo-
tagged photos or in-situ surveys). Based on these 3D-building models
we introduce a method for classification of physical built-up categories
of Arrival Cities. Methodologically we extend established approaches
on the measurement of patterns of the built environment from EO data
(based on Taubenböck & Kraff, 2014). In consequence, this study relies
on a data set which has been produced in a consistent and comparable
way, allowing a single methodological logic and is thus unbiased from
possible inconsistencies that usually occur, if multiple input data across
countries are applied. With it, we aim to add to the current body of
literature for a better empirical documentation of the settlement
morphologies of Arrival Cities.
Specifically, we aim to answer the following research questions: (1)
Which physical morphologic settlement categories of Arrival Cities can
be distinguished? (2) Which similarities and differences between
building morphologies of Arrival Cities exist within a city, a country or
a continent?
The remainder of the work is structured as follows: Section 2briefly
presents the relevance of this topic on the political development agenda
and reviews the state of the art from an urban remote sensing per-
spective. Section 3introduces the ‘Arrival City’as conceptual frame-
work for using a physical approach towards classifying and character-
izing morphologies of the urban poor. Section 4presents the
methodology of the literature survey, the selection process of re-
presentative Arrival Cities, the classification of 3D building models, and
finally the analysis of spatial patterns and their categorization. In sec-
tion 5the results on the measured spectrum of morphologies in Arrival
Cites and the resulting categories are presented. This is followed by a
discussion in section 6, where we also try to evaluate which influence
determinants such as the topographic situation or pre-existing patterns
have on urban morphology. Section 7concludes with a perspective.
2. Background and state of the art
The intergovernmental agreement on the Sustainable Development
Goals (SDGs) acts as the post 2015 Development Agenda (successor to
the Millennium Development Goals) (UN, 2015). To end poverty, en-
sure healthy lives, provide access to quality education, water and sa-
nitation, to build sustainable cities and communities, among others are
goals directly relating to challenges for the urban poor. However, the
demand for ‘sustainable data for sustainable development’or, in other
words, improved data availability, quality, consistency, timeliness and
disaggregation is often not met (UN, 2015).
2.1. Approaches towards assessment of the urban poor
A number of statistics underpin the urgent need for more extensive
empirical knowledge on the places of the urban poor. 25% of the global
urban population (which is almost 1 billion people) live in slums or
informal settlements (UN-Habitat, 2015). The builders of informal
housing have become the largest builders of housing in the world
(Tiwari, 2007), and thus, they are creating the cities of tomorrow. Davis
(2007) estimates 200,000 slums globally, not to mention the unknown
higher quantity of Arrival Cities. These often neglected parts of the city
are of crucial importance due to their arrival functions providing access
to the labour markets, education, etc. In Mumbai, for example, 65% of
the workforce is employed in the informal sector (Sudjic, 2010)orin
Mexico City 60% of construction is informally done (Burdett & Sudjic,
2007).
Although we observe an increasing quantity and variety of pub-
lications on this topic of urban poverty, these still lack consistent con-
ceptual understandings across or even within disciplines, agreed
methods of measurement or empirical data necessary. This leads to, as
Satterthwaite (2003) proclaimed “there is no lack of nonsense statistics on
levels of urban poverty”, or the general statement in the World Migration
Report (2015) that we have a massive lack of basic data about urban
poverty.
From a spatial point of view, the coverage of data on areas of the
urban poor varies significantly across the globe; especially in the Global
South large data gaps exist. In consequence, a systematic quantitative
spatial documentation of morphologic forms is absent. With the recent
massive increasing availability of EO sensors, remote sensing data have
become crucial for spatially capturing urban inequality. Freed from any
administrative boundaries, the data are consistent and (in theory as e.g.
data costs are still a restriction) available globally. Nevertheless, with
only 87 key publications identified to date in the field of remote sensing
(Kuffer, Pfeffer & Sliuzas, 2016), we are still far away from a global
inventory of Arrival Cities or a systematic classification of physical
characteristics of these places. However, we must be aware that using
the morphologic structures as proxy for identifying the social group of
the urban poor is simplifying the spatial and social complexity (e.g.
Wurm & Taubenböck, 2018). In consequence, the remote sensing
community conceptualizes poverty by features describing the mor-
phologic urban appearance (Sliuzas et al., 2008). In a generic slum
ontology indicators are conceptualized allowing a localization of set-
tlements of the urban poor from EO-data (Kohli, Sliuzas, Kerle, & Stein,
2012). However, as slums are a relative concept (Gilbert, 2007), a
physical approach to mapping these areas still lacks an internationally
agreed concept as well as systematic empirical documentation. As a
result studies using remote sensing data still lead to incomparable data
sets across studies and study sites due to conceptual differences and the
relativity of the target classes.
2.2. Mapping the urban poor with EO-data
From a remote sensing perspective, recent work predominantly fo-
cused on the development of methodologies for automatic classification
of slum areas. Based on morphologic assumptions defining these areas
of the urban poor, considerable methodological development has been
achieved to automatically classify locations and extents of slums from
optical (e.g. Hofmann, Strobl, Blaschke, & Kux, 2008; Kuffer & Barros,
2011; Kuffer, Pfeffer, Sliuzas, & Baud, 2016; Shekhar, 2012) as well as
radar (e.g. Stasolla & Gamba, 2008; Wurm, Taubenböck, Weigand, &
Schmitt, 2017) data. The developed approaches rely on physical fea-
tures characterizing the built environment by indicators such as high
heterogeneity in building alignments (e.g. Jain, 2007; Owen & Wong,
2013), irregular street networks (e.g. Gueguen, 2014; Niebergall, Loew,
& Mauser, 2008), small building sizes (e.g. Baud, Kuffer, Pfeffer,
Sliuzas, & Karuppannan, 2010; Graesser et al., 2012) or high building
densities (e.g. Kit, Lüdeke, & Reckien, 2012; Taubenböck & Kraff, 2014)
to differentiate these housing areas from formal ones. Studies use
spectral, spatial and textural features and combinations of them (e.g.
Engstrom et al., 2015; Graesser et al., 2012; Kit & Lüdeke, 2013;
Sandborn & Engstrom, 2016; Wurm, Weigand, Schmitt, Geiß, &
Taubenböck, 2017) for operationalizing the observed built-up features
by image features. In this regard, Kohli, Stein and Sliuzas (2016) pre-
sent related uncertainties in image interpretation.
Beyond, the combination of remote sensing with other data sources
for detecting (and characterizing) areas of the urban poor is an emer-
ging research field. The interplay of EO-data with census information
proves correlations of image features with socio-economic parameters
of the areas (e.g. Duque, Patino, Ruiz, & Pardo-Pascual, 2015; Sandborn
& Engstrom, 2016; Taubenböck et al., 2009). Wurm and Taubenböck
(2018) prove that morphologic slums classified from EO-data allow the
localization of urban poor with high accuracies; however, spatial
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
152

differences between morphologic and census slums also reveal short-
comings of the physical approach towards completeness. Applications
of EO-based morphologic slum classifications are e.g. the assessment of
slum populations (Kit, Lüdeke & Reckien, 2013; Taubenböck & Wurm,
2015) or the combination with geotagged data from social media
identifying most areas of urban poor being digital deserts (Klotz, Wurm,
Zhu, & Taubenböck, 2017).
At the spatial level of 3D building models, studies documenting the
built environment are very scarce. This is due to the complexity of slum
morphologies in combination with the low availability of highest re-
solution (< 1 m) digital surface models. Few studies measure the
mentioned physical features of slums in a quantitative sense and com-
pare morphologies within (Taubenböck & Kraff, 2014) and across cities
(Taubenböck & Kraff, 2015). Studies also acknowledge that even within
one city, physical morphologies, roof appearances as well as types of
slums vary significantly (Kuffer, Pfeffer, Sliuzas, Baud, & Maarseveen,
2017).
Summarizing, although remote sensing is a crucial data source for
detecting and characterizing areas of the urban poor, significant
knowledge gaps exist: 1) remote sensing reduces the social group of the
urban poor to slum areas; 2) classification algorithms remain proof of
concept for limited test sites failing to provide consistent large area
(continental scale) classifications of slums with high accuracies; 3)
documentation of the variability of morphologic forms of the urban
poor at 3D city model resolutions remain scarce; and 4) an inventory of
morphological types as conceptual foundation is inexistent due to data
gaps. In this paper, we aim with our comprehensive empirical baseline
study at reducing the data gaps regarding physical morphologic cate-
gories of Arrival cities across the globe and we aim at providing the
knowledge for algorithm development capturing these areas in EO-data.
3. Conceptualization of this study –the Arrival City
The challenge of a global characterization of physical appearances
of settlements titled ‘slum’,‘informal’,‘squatter’,‘spontaneous’,‘ghetto’,
‘illegal’or ‘irregular’or described by local names such as ‘favela’,‘bi-
donville’,‘township’or ‘gecekondu’,among many others, starts with
terminology and related conceptualizations. Given the attention in lit-
erature, one expects the evolution of a clear and consistent con-
ceptualization and method of measurement; however, these terms do
not base on a standard definition, they are conceptually neither distinct
nor consistent (Ghani & Ranbur, 2015).
By comparing the often used terms slum versus informal settlement,
the conceptual fuzziness becomes obvious. UN-Habitat (2003) defines
slums by using qualitative measures based on the so called “shelter
deprivation”indicators such as the lack of durable housing and tenure
security, overcrowding, and limited access to clean water and accep-
table sanitation. The definition of informal settlements differs (although
the term is often applied synonymously; in consequence, this leads to an
imprecise usage in literature): Informal settlements refer to those areas
that developed through unauthorized occupation of land outside a
legal, regulatory, planned and professional framework (e.g. Bähr &
Mertins, 2002; Huchzermeyer & Karam, 2006). The consequence is that
slums refer to highly precarious living conditions whereas informal
settlements, in contrast, are not exclusive to these living conditions as
the concept relates to the legal jurisdiction. Thus, informality may also
refer to better quality of construction and living conditions. Other as-
sociated problems arise e.g. as slums are relative in their conceptual
meaning: What is considered to be a slum in one country may be re-
garded as perfectly acceptable accommodation in another (Gilbert,
2007). In consequence, slums cannot be defined safely (or measured in
absolute terms) in any universally acceptable way; in the case of in-
formal settlements, their genesis may have occurred outside formal
means, but over time got formalized (e.g. by slum upgrading pro-
grammes) making it difficult for unambiguous classification. Ambiguity
on these areas relates to the common understanding that informal (and
also formal) forms of settlements rarely occur in any pure form; hybrid
conditions (e.g. complex compositions of organic building layouts in-
terwoven with structured planned forms) are the norm (Werthmann &
Bridger, 2016).
Against these issues on terminology and related conceptual com-
plexity, we base this study on the term and concept of the ‘Arrival City’.
The term has been introduced by Saunders (2010), capturing places
with the main function of arrival to the city. Semantically the term
'arrival' refers to the relocation of rural populations to urban environ-
ments. However, conceptually the term is not just meant as a sole
physical arrival functioning as a mere place for living and working; it is
most importantly meant as a place of comparatively cheap living con-
ditions that opens up the possibility to become part of the urban society.
In consequence, this concept integrates all places which provide low-
cost access to ‘arrival functions’(living, working, education, cultural,
etc.) in cities, which morphologically span from spontaneous, pre-
carious shacks built overnight to squatting of informal or formal land
and/or structures, to (once respectable) housing affected by deteriora-
tion. The concept of the Arrival City is thus embracing the conceptual
differences of the other mentioned terms.
But why is this concept of the Arrival City necessary for this study?
The literature survey (cf. section 4.1.; appendix 1) reveals that an un-
ambiguous conclusion on the status of a study site regarding in-
formality, security of tenure, access to sanitation, etc. is not always
possible; furthermore, as hybrid forms are the norm a discrete classi-
fication can obscure reality. A conclusion on the functioning as Arrival
City is, in turn, more straight-forward and unambiguous. Another issue
is that, as mentioned above, various popular terms such as ‘slum’or
‘informality’are very inconsistently applied in different studies (e.g.
Kuffer, Pfeffer & Sliuzas, 2016); a comparison of morphological pat-
terns is at least at risk to be conceptually illegitimate. The lack of ter-
minological consistency especially in literature dealing with physical
appearances (mostly using remote sensing data) results in ontologies
and classifications which remain conceptually vague, inconsistent and
incomparable.
With this umbrella concept of the Arrival City we overcome these
issues and aim at the following specific aims: 1) with the increasing
availability of EO data, this data source recently became essential for
classification of these areas. However, with respect to the conceptual
ambiguity, the capability to measure inequality in cities from space by a
spatial approach needs a systematic catalogue of morphological forms.
Thus, we aim to bring clarity to the variability of morphological types
of Arrival Cities. 2) The existing different concepts lead to incomparable
data across nations or continents. With the morphological catalogue,
we aim at a distinct conceptual framework; in terms of remote sensing,
this aims at baseline information for algorithm development by mor-
phologically defined target classes.
4. Data and methodology
This section is structured into two parts: The first part (4.1.) in-
troduces the strategy of study sites selection aiming at a representative
global sample of Arrival Cities. The second part (4.2.) introduces the
spatial concept for classification of physical appearances of the settle-
ment structures, the quantitative spatial analysis of the morphologic
patterns and the categorization strategy.
4.1. Literature survey and selection of study sites
A comprehensive global knowledge repository about Arrival Cities
to draw from does not exist. Beyond, an exhaustive compilation of the
entire variability of the physical appearances of settlements is, at a
resolution of individual buildings, still an utopia. Reasons are that in-
formal settlements are mostly absent in official plans, not all existing
physical types of informality have been reported or systematized and
more generally, availability of adequate highest resolution geodata
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
153

especially in slum-like structures is still scarce. This means a high un-
certainty on amount, spatial distribution and types of Arrival Cities
exists. In consequence, we conduct a systematic literature survey to
compile a large variety of documented Arrival Cities across the globe to
provide a representative sample.
For the systematic search we use general search terms such as ‘ar-
rival city’,‘informal settlement’,‘slum’,‘squatter’,‘shanty town’and
‘ghetto’as well as synonyms in the respective local context such as
‘favela’,‘gecekondu’, among many others. Our search relies on common
search engines such as scopus or google scholar (cf. all criterions ap-
plied for the systematic search are presented in Table A-2 in the ap-
pendix). Yet, not any documented and identified Arrival City in lit-
erature is intended to be listed; instead we aim at selecting
representative study sites by reason of different attributes:
(a) we select Arrival Cities documented in literature
(b) we select Arrival Cities from different cultural areas and continents
for a representative global geographic distribution.
A main challenge arises from the necessity for 3D building models
for a structural spatial analysis (cp. spatial concept in 4.2.). The selec-
tion process therefore is guided by the limitation that every selected
Arrival City needs an extensive classification based on available very
high resolution (VHR) EO-data. Based on these considerations we
choose Arrival Cities for a quantitative spatial analysis by reason of the
following attributes:
c) we select Arrival Cities consisting of a spatial extent formed by a
significant group of buildings (> 1000), not by one individual or
few buildings.
d) we select a generally even distribution of Arrival Cities in different
topographic locations; this allows specific analytic questions whe-
ther the physical appearance shows differences due to topography.
e) we select for few examples more than one Arrival City within one
city; this allows specific analytical questions whether the physical
appearance reveals intra-urban differences.
f) we select Arrival Cities originating from time periods between 2010
and 2016.
In the literature survey many morphologic forms of Arrival Cities
are identified which do not fit into our spatial concept; e.g. these are
areas with untypical spatial extents that dissent to a significant amount
of buildings or there are forms that are less conform to ‘settlements’
with regard to accommodation. For a more comprehensive picture we
additionally provide a qualitative description of these morphologies for
amore comprehensive presentative of physical forms of living.
4.2. Classification and characterization of settlement morphologies
What defines the appearance of a settlement? And how can it be
measured and characterized? In general, types and conditions of in-
dividual buildings as well as the spatial alignment of these buildings define
the urban spatial structure. In consequence, we aim at measuring the
appearance of a settlement by the spatial pattern buildings create (1)
and the (building) morphology (2):
(1) For characterizing spatial patterns of settlements we apply three
features: building density (in 2D), orientation of buildings and hetero-
geneity index (relating to the variance of density within the area of
interest).
(2) For characterizing the morphology of the individual buildings we
use two features: size (ground floor) and height (number of stories).
The condition of the building is disregarded here, as this parameter
is difficult to assess using EO-data.
In the workflow we use three spatial scales of analysis, from the
smallest to the largest: Building level (a), block level (b), district level (c)
(Fig. 2).
Fig. 2. Hierarchical structure of spatial scales –building, block and district level –for the morphologic settlement analysis and the workflow for calculating the five structural variables
and the final categorization of Arrival Cities.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
154

a) Building level: The building level refers to the highest spatial detail
–the individual buildings (Fig. 2). We derive the necessary high
detailed geoinformation from very high resolution optical satellite
data in combination with Google Street view information, geor-
eferenced photos and partially in-situ observations. As structural
complexity in Arrival Cities does mostly not allow for fully auto-
matic delineation of individual building footprints with acceptable
accuracies from EO-data, we derive the complex settlement struc-
tures using the cognitive perception of an experienced image ana-
lyst. Employing a standardized digitization protocol a consistent
data base is derived at building level: Using EO-data having geo-
metric resolutions of 1 m and better (e.g. QuickBird, WorldView)
and a consistent scale of 1:1000 for image analysis, each building is
digitized and represented by a single polygon. The polygons contain
several vertices representing real shapes of ground floors even for
more complex buildings than rectangle structures. We are aware
that in certain areas the complexity of dense building patterns and
respective roof structures is even outdrawing the visual capabilities
to distinguish individual buildings in VHR optical EO-data; thus it
might lead in certain cases to a generalized derivation of a polygon
representing a mixture of houses instead of an individual un-
ambiguous polygon per building. However, as Sliuzas et al., 2008
remark, the visual interpretation still offers the best capability for
deriving these complex structures.
From the building classification, the individual building sizes and
orientations are calculated. Beyond this, building heights (as number of
floors) are attributed to every individual building. To do so, we in-
corporate in-situ information on building heights from field surveys
(done for the samples in Mumbai, Izmir, Istanbul, Sao Paulo, Athens,
Bucharest, Berlin, Nairobi and Cape Town), we employ spectral and
spatial characteristics in VHR EO-data (we use estimations by the in-
terpreter such as shadows or building patterns allowing an assessment
of building height) and analyse georeferenced photos in online plat-
forms such as GoogleEarth or, if available, Google Street view in-
formation (where we count the number of floors). In a previous study
the building heights assessed using this approach resulted in an accu-
racy of 91,9% (Taubenböck & Kraff, 2014). Based on the parameters
building sizes and heights we achieve 3-D city models in level of detail
1 (LoD1). All variables measuring the appearance of settlements relate
to these 3D city models.
b) Block Level: The city block is a spatial entity combining the in-
dividual elements (buildings) into spatial aggregates aiming to
capture spatial patterns of the settlements on a meso scale (Fig. 2).
To do so, the close meshed street network is used to define city
blocks; or, if in organic (informal) settlement structures streets are
absent, visually identifiable pathways or obvious structural change-
overs are introduced manually for the sub-division of space. As ex-
ample, if a dense settlement area turns abruptly into an open space
without a street as separating line, we introduce one to provide
spatial entities capturing high and low density without mixed areas
with blurring effects. In general, the city blocks are spatially sub-
dividing the district level.
At this block level, we calculate five features to spatially measure
the built environment of a settlement: we calculate the building density
as the ratio between the sums of all cumulated building ground floors to
the respective block unit area (eq. (1)).
=∑=
BBF
A
dn
n
1n
R
Ru
u(1)
A = area; B
d
= building density; BF = extent of building footprint;
R
u
= reference unit; n = number of buildings;
We derive the orientation of buildings as proxy for measuring the
complexity of the alignment of buildings. We assume unplanned areas
are by nature more irregular than planned. To do so, we calculate the
individual orientation of each building husing the longitudinal side.
Subsequently, the nearest spatial neighbor of each building is detected
and the difference in main orientation O
Δ
is calculated as difference in
angular degree (eq. (2)).
=− = = ≤ ≤
°°
OOO b O
Δ
1, ..., B; h 1, ..., H ; 0 90
bh bh N bh bh() b (2)
where Bis the total number of building blocks, H
b
is the number of
buildings per block b,
O
bh is the orientation to geographic north and N
(bh) describes the one nearest neighbor of each building within the
block.
We assume a geometric order of building pairs if orientation dif-
ferences are close to 0 or 90°. In contrary, we assume complex align-
ments closer to orientation differences of 45°. We convert this into a
continuous index I
bh
of orientation complexity ranging from 0 (=geo-
metric) to 1 (=highest degree of geometric chaos) for every building
pair (eq. (3)).
=− − × +IOΔ45
1
45
1
bh bh
(3)
We aggregate all individual index values of building pairs to de-
termine the median alignment index for the respective blocks (eq. (4)).
==…=Imed b{I h 1, ,H }| 1,....,
B
bbh b
(4)
where H
b
is the number of buildings in the block b.
As third feature, we measure the morphologic heterogeneity of the
pattern over space. The heterogeneity index I
h
quantifies differences in
building densities of a block with all adjacent neighboring blocks.
Therefore, absolute building density differences from a center block to
all adjacent neighboring blocks are calculated. The individual density
differences are added in absolute values (as the subtraction of density
values of the center block with the density value of an adjacent block
may result either in a positive or a negative value). The respective
quantity of norm changes is noted also. This sum of norm differences
gets divided by the quantity of neighboring blocks for normalization
leading to a nondimensional number. The number of norm changes will
be multiplied with the sum of building density differences; if a block has
no norm changes, it will always be added to 1. With increasing differ-
ences of density values in adjacent blocks, the index indicates a higher
heterogeneity of a pattern (eq. (5)) (for more details see also
Taubenböck & Kraff, 2014).
=⎛
⎝
⎜∑−⎞
⎠
⎟∗
=
IN
C
|bd bd
k
hi
k
1ci
(5)
I
h
= heterogeneity index; k = quantity of adjacent neighboring blocks;
bd
c
= built-up density value of the center block; bd = built-up density
value of an adjacent block, NC = quantity of norm changes.
As fourth feature, we calculate the building sizes as the average of all
building ground floors in the respective block unit area (eq. (6)).
=∑=
BBF
n
sn
n
1n
R
Ru
u(6)
B
s
= average building size.
And, as fifth feature, we calculate the building heights as the average
of all building floors within the respective block unit area (eq. (7)).
=∑=
BA B
n
hn
n
1h
R
Ru
n
u(7)
B
h
= building height; BA
h
= average building height.
c) District level: As main spatial level of analysis we employ the dis-
trict level (Fig. 2). The district level represents the entire area of one
Arrival City (or at least for the parts where geoinformation have
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
155

been derived). This unit functions as one, consistent spatial level for
the aggregated morphologic settlement analysis. Here, the varia-
bility of the five features at block level are presented by providing
the median (for building heights we apply the mean, as the limited
number of classes and the often dominant low building heights do
not allow for a precise differentiation using the median) and the
data distributions are presented in box plots.
4.3. Categorization of settlement morphologies
For the categorization of the measured settlement morphologies we
quantify deviations from measured spatial features against an expected
(model) value. The expected (model) value represents the measured
maxima per feature across all Arrival Cities. For every one of the five
spatial features we state a hypothesis. We let ourselves guide from the
common usages in literature for a subtype of Arrival Cities –morpho-
logic slums –which we assume to have patterns of highest morphologic
complexity:
1) We expect high building densities as open space in cities is limited
and precious, population pressure is high and planning regulations
are absent, which leads to a minimization of open public space.
2) We expect non-geometric, irregular orientations of buildings as
planning regulations are absent, regular street pattern not given,
individual decision form random lay-outs, and –in certain cases–
adjustments to complex topographic situations may lead to non-
regular lay-outs.
3) We expect high structural homogeneity of the building patterns as
full utilization of space is targeted as space is limited.
4) We expect shelters of small sizes as available land is limited and
precious, population pressure is high and financial capabilities of
the dwellers are low.
5) We expect low building heights as financial capabilities of the
dwellers are low and building materials are limited.
The virtual combination of measured maxima per feature generates
atheoretical ideal type morphologic slum. This virtual combination is, of
course, not existent in reality as the maxima per feature result from
different Arrival Cities; however, it marks the virtual end of measured
real-world structures for calibration in the analysis.
We categorize a morphologic settlement type index value I
m
as devia-
tion from the theoretical ideal type morphologic slum. To do so, we find
maximum and minimum values per feature of the calculated medians
(mean for building heights). By minimum-maximum normalization [0,
1] (eq. (8)), we adjust heterogeneous values measured to a notionally
common scale per variable.
x′:= (x−x
min
)/(x
max
−x
min
) (8)
x
min
is the minimal median value per feature; x
max
is the maximum
median value per feature.
The I
m
results from the cumulative deviation from the theoretical
ideal type morphologic slum (eq. (9)).
I
m
=x′(d) + x′(o) + (1−x′(h
i
)) + (1−x′(s)) + (1−x′(h)) (9)
I
m
is the morphologic index, d = density; o = orientation;
h
i
= heterogeneity index; s = building size; h = building height.
With it, the measured physical appearances of Arrival Cities can be
classified along a continuous scale for categorization. However, as
auxiliary means for semantic description we apply discrete categories
using equal distances among groups.
Beyond, we classify additional categories by descriptive analysis for
the Arrival Cities identified in the literature survey, but not fulfilling the
spatial concept of analysis.
5. Results
This section is organized as follows: First, the selected Arrival Cities
used for the spatial analysis of settlements are presented and related
background information from the literature survey is provided (5.1.).
Second, the physical characteristics of the selected Arrival Cities are
presented (5.2.). Third, we introduce the categorization of the Arrival
Cities (5.3.). In this part we relate the categories to different location
based observations, i.e. whether categories have obvious similarities
within one city or country. Fourth, we present other types of Arrival
Cities which do not fit into the spatial concept of analysis but are re-
levant forms (5.4.).
5.1. Global representatives for Arrival Cities
Based on the literature survey we select 44 Arrival Cities across the
globe. With 10 Arrival Cities from Africa,9 from America,15 from Asia
and 10 from Europe, a global distribution is realized. Within the con-
tinents a spatial distribution is carried out (e.g. for Asia we chose
Arrival Cities from western (e.g. Turkey), central (e.g. Iran, India), and
eastern parts (Mongolia, China, The Philippines)). Beyond, the sample
also contains different Arrival Cities within the same country (e.g. for
Brazil, China, France, Turkey and USA) as well as different Arrival
Cities within sample cities (e.g. Cairo, Cape Town, Mumbai and Dhaka).
Fig. 3 illustrates the spatial distribution of the selected samples and
visualizes spatial appearances of selected settlement structures in very
high resolution optical EO-data. It becomes obvious that building
morphologies being the home of the urban poor feature a large varia-
bility.
In the appendix (Table A-1) comprehensive background information
on all 44 selected Arrival Cities are introduced: locations, names, sizes
of the areas of interest, the number of buildings classified from Earth
observation data. Furthermore, the table gives systematic background
information on the topographic situation, descriptive information on
housing types and materials, access to infrastructure, legal status, if
possible, and the estimated population. The reference from literature
which defines the selected sample as Arrival City complements the
table.
5.2. Physical characteristics of Arrival Cities
Under the conceptual umbrella of the Arrival City we unite terms
such as slums, informal settlements and the like. These terms implicitly
carry the idea of a uniform urban type including morphologic simi-
larity. The reported physical appearances of these settlements are
mostly of qualitative nature and do not allow for an objective, quan-
titative generalization. Other studies using EO-data take conceptually a
certain morphologic similarity as a given basis.
In general, the classifications of Arrival Cities from EO-data pre-
sented in ground figure plans (Fig. 4) illustrate that morphologic si-
milarity cannot be taken for granted. The visualized building ground
floors and their patterns feature variabilities, but also similarities. We
find that, as many descriptive studies suggested, high building den-
sities, complex, organic patterns and small building sizes are features
applying to many of the Arrival Cities. However, a closer inspection
reveals that it is not as simple as that: We also find significant lower
building densities, large building sizes, lay-outs with a geometric or-
ganization, in all kinds of combinations. As examples, we find in highly
dense areas both, complex (e.g. number 06 in Fig. 4) and ordered (e.g.
13) alignments of buildings; we find large (e.g. 37) vs. small buildings
(e.g. 31). We also find highest building densities (e.g. 05) vs. com-
paratively low densities (e.g. 44).
The visual inspection of the ground figure plans does not allow for a
quantitative, resilient analysis of the building patterns and
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
156

Fig. 3. Selected Arrival Cities across the globe from the literature survey and illustration of the appearance of different morphologic forms of the built environment in EO-data.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
157

Fig. 4. Ground figure plans of the 44 selected Arrival Cities.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
158

Fig. 4. (continued)
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
159

morphologies. Fig. 5 presents and contrasts the five features defining
these urban morphologies at block level–building density,building or-
ientation and heterogeneity index representing the patterns and building
size and height representing the building morphologies –in form of
boxplots. The 44 Arrival Cities are grouped based on their continental
location. Beyond, each variable is ordered in decreasing or respectively
Fig. 5. Boxplots illustrating five variables defining urban morphologies for all 44 selected Arrival Cities.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
160

increasing manner of the median (mean for building heights) for the
particular feature.
In general, we find that a simplistic approach towards the mor-
phology of the living areas of urban poor obscures reality. On the one
hand, the boxplots for almost all features and study sites reveal that we
are dealing with hybrid forms of building patterns and morphologies, as
they feature more or less variance within and across Arrival Cities. First,
we detect that a large variance within continents appears (e.g. we find
on every continent Arrival Cities with building densities of 70% and
more to below 40%). Second,wefind large varieties of morphologic
forms within one country (e.g. with building heights of 1.57 stores on
average for Gülsuyu-Gülensu, Istanbul vs. 4.59 stores in Kadifekale,
Izmir). Third,wefind large varieties of morphologic forms within one
city (e.g. building sizes of Eldarb el-Ahmar in Cairo are in median with
131 m
2
more than twice as large as the ones in Manshiet Nasser with
60.5 m
2
). On the other hand, the boxplots also reveal that although these
morphologically hybrid forms are identified, the stated hypothesis on
typical morphologic features (cf. section 4.3) can be confirmed to a
certain degree: First, we measure that high building densities are a
characteristic feature of Arrival Cities. Although building densities
fluctuate across the globe, 21 out of the 44 Arrival Cities feature den-
sities higher than 60% (26 with densities higher than 50%). This re-
markable intense utilization of space can be benchmarked when we
relate these values to other parts in cities. As examples, the medians of
building densities are significantly lower in central areas of mega cities
(e.g. in a circle of a 10 km radius around the center of London, UK the
density is 30% or in Paris, France it is 34%). They are also significantly
lower for planned cities (Chandigarh, India 35%; Lingang, China 24%),
large housing estates (e.g. Neuperlach, Munich 18%; Gropiusstadt,
Berlin 16%) or residential suburbs (e.g. Sun City, USA 24%, garden city
Letchworth, UK 15%). Within our (incomplete) sample for comparison
of building densities to Arrival Cities, historical European city centers
reveal densities closest to Arrival Cities (Hamburg, Germany 46%;
Dortmund, Germany 46%; Munich, Germany 52%). Second, we measure
that the hypotheses for the heterogeneity index (29 Arrival Cities below
the value of 15, which is comparatively homogeneous), the building
sizes (half of the study sites have ground floors below 60 m
2
) and
heights (24 Arrival Cities are not higher than 2 floors in average) can
largely be confirmed. In contrast, for the building orientation a clear
trend towards complex alignments is not existent (12 Arrival Cities are
below 0.2, what indicates a relatively regular orientation; 16 Arrival
Cities feature values between 0.2 and 0.4, and 16 Arrival Cities are
above the value 0.4).
5.3. Morphologic categorization of Arrival Cities
For the morphologic categorization of Arrival Cities we transfer the
features into a common scale by the minimum-maximum normal-
ization. The cumulative combination of all normalized variables results
in the morphologic settlement type index. The index allows classification
of the different Arrival Cities into morphologic categories.
As we base our categorization on relative deviation from the
Fig. 6. Categorization based on the developed morphologic settlement type index value I
m
for all 44 Arrival Cities.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
161

assumptions we stated per variable, the morphologic type fulfilling the
expectations of all variables with 100% is a virtual combination of
maxima measured for all 44 Arrival Cities (class A
t
in Fig. 6). The
maxima are Islambagh, Dhaka for building density (Median = 90.2%),
the refugee camp Afgooye for building orientation (0.27), Bharat
Nagar, Mumbai for the heterogeneity index (5.2), the north cemetery,
Manila for building size (7.7 m
2
) and West Adams, Los Angeles for
building height (1.5 floors). The large range of morphologic categories
becomes tangible when we compare these values with the ones mea-
sured with maximum deviation (class D
t
in Fig. 6). This is the virtual
combination of the refugee camp Al-Zataari for building density
(Median = 24.3%), North Philadelphia, Philadelphia for building or-
ientation (0.0), Khayelitsha (Victoria Merge), Cape Town for the het-
erogeneity index (50.8), Les Pyramide, Évry for building size
(685.9 m
2
) and Baishizhou, Shenzhen for building height (7.7 floors).
They mark the unexpected virtual end of morphologic categories (cf.
Table 1).
In general, we find there is no homogeneous morphological global
everywhere of Arrival Cities. The social group of urban poor trying to
get access to urban societies and functions live in very different struc-
tural patterns and building morphologies. Referring to the first specific
research questions stated in the introduction, we classify three main
categories and three respective transitional forms. They stretch from slum
(Cat. A; threshold > 3.75) and slum-like (Cat. AB; range 3.75–3.25)
morphologies to mixed unstructured-structured neighborhoods (Cat. B;
range 3.25–2.75) and mixed structured-unstructured neighborhoods (Cat.
BC; range 2.75–2.25) even to structured (Cat. C; range 2.25–1.75) and
formally planned (Cat. CD; range 1.75–1.25) areas. In general, the ap-
plied continuous scale allows an unambiguous assessment of the mor-
phologic conditions. However, for a more general description and no-
menclature we use the mentioned thresholds of equal-distance. Table 1
introduces the categories, describes the measured physical features and
lists samples and the resulting morphologic settlement type index va-
lues.
Regarding the second specific research question stated in the in-
troduction we find similarities and differences between building
morphologies of Arrival Cities: Within a single city we find that places of
urban poor do not necessarily reflect similar morphologies. In Cape
Town the morphologic differences reach from the (informal) slum-like
structures of Griffiths Mxenge (Cat. AB; I
m
of 3.60) to the planned
township of Mandela Park (Cat. C; I
m
of 2.32); in Dhaka they reach from
mature downtown slums where continuously re-densification processes
Table 1
Morphologic categories of Arrival Cities.
Cat. Terminology Description Samples & morphologic index
A
t
Theoretical ideal type
morphologic slum
This theoretical morphology reflects the combination of the extrema
per physical feature measured from the 44 samples across the globe.
Density: Islambagh, Dhaka (90.2%); orientation: Afgooye (0.27);
heterogeneity index: Bharat Nagar, Mumbai (5.2); size: north
cemetery, Manila (7.7 m
2
); height, West Adams, L.A. (1.5 floors);
(morphologic index =5.0).
A Morphologic slum The morphology measured in one real Arrival City corresponds to the
greatest possible extent with the physical assumptions in our spatial
concept as well as with the suggested ontologies and qualitative
descriptions in literature. Small makeshift shelters are huddled
together in most complex alignments.
Kibera, Nairobi (4.36); Bharat Nagar, Mumbai (4.19); Tondo,
Manila (4.19); Makoko, Lagos (4.08); Petare, Caracas (4.03);
Dharavi, Mumbai (4.01);
Santosh Nagar, Mumbai (3.95); Magdalena Contreras, Mexico City
(3.93); Islambagh, Dhaka (3.82); Hutong, Beijing (3.81); Afgooye
(3.78);
AB Slum-like morphology The morphology features deviations from the measured extrema or
the common assumptions in at least one of the five physical features.
However, the dominant physical appearance is a very dense,
complex pattern of deprived building types.
Lo Prado, Santiago de Chile (3.62); Griffiths Mxenge, Cape Town
(3.60); Nadezhda, Sliven (3.58); Turano, Rio de Janeiro (3.56);
Cova de Moura, Lisboa (3.50); Victoria Merge, Cape Town (3.50);
Gülsuyu-Gülensu, Istanbul (3.43); North Cemetery, Manila (3.41);
Manshiet Nasser, Cairo (3.39); Independencia, Lima (3.36);
Paraisópolis, Sao Paulo (3.30); Imbaba, Cairo (3.30); Khoroo 9,
Ulan Bator (3.30); Refugee Camp, Al-Zataari (3.30); Ezbet el-
Hagana, Cairo (3.29); West Adams, L. A. (3.29); Eldarb el Ahmar,
Cairo (3.25);
B Mixed unstructured-
structured neighborhoods
The morphology features significant deviations from the measured
extrema or the common assumptions of slum morphology in more
than one of the five physical features; it contains mixed forms still by
trend closer to slum morphology than to structured, formal
neighbourhoods: Forms include further developed once slum-like
morphologies (e.g. increase in building heights), run-down, deprived
(and once higher quality) building blocks, infiltration of shelters into
existing residential structures, or converting of shelter usages for
urban poor.
Tei Toboc, Bucharest (3.25); Eslamsahar, Teheran (3.17); Agios
Panteleimonas, Athen (3.07);
Canada Real, Madrid (2.99); Kadifekale, Izmir (2.97); Belleville,
Paris (2.80); Karaagac-Altigagac, Ankara (2.78);
BC Mixed structured-
unstructured neighborhoods
The morphology features significant deviations from the measured
morphologic slums and slum-like morphologies as well as from the
related common assumptions. The morphology combines typical
features of structured (e.g. geometric alignments, frequent spatial
transition of buildings and open spaces) und unstructured
neighbourhoods. The morphology is by trend closer to structured,
formal neighbourhoods
Kadamtali, Dhaka (2.73); Baishizhou, Shenzhen (2.57);
Molenbeek, Brussels (2,43); Mandela Park, Cape Town (2.32);
Kreuzberg, Berlin (2.26);
C Transition to structured
character of neighborhoods
The morphology combines typical features of planned, structured
neighborhoods such as regular alignments or lower densities with
few slum-like features.
North Philadelphia, Philadelphia (2,17); Tower Hamlets, London
(2.16);
CD Formal, structured
neighborhoods
The morphology provides typical features of planned, formal,
structured neighborhoods: low densities, geometric alignments, large
and high buildings.
Le Pyramide, Évry (1.53). Bronx, New York City (1.52);
D
t
Theoretical ideal type formal,
structured neighborhood
This theoretical morphology reflects the combination of the minima
per physical feature measured from the 44 samples across the globe.
Density: Al-Zataari (24.3%); orientation: North Philadelphia,
Philadelphia (0.00); heterogeneity index: Victoria Merge, Cape
Town (50.8); size: Le Pyramide, Évry (685.9m
2
); height:
Baishizhou, Shenzhen (7.7 floors);
(morphologic index =1.0).
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
162

occurred over time (Islambagh, Cat. A; I
m
of 3.82) to more recent, more
peripheral informal developments where full utilization of space still
lies ahead (Kadamtali; Cat. BC; I
m
of 2.73). However, we also observe at
the same time that structures of the same category are reproduced in
very similar manner within one city (e.g. in Cape Town the areas of
Griffiths Mxenge and Victoria Merge are both Cat. AB with I
m
of 3.60
and 3.50; or in Cairo all four Arrival Cities are classified as Cat. AB).
Logically, also within a single country places of urban poor do not
necessarily reflect similar morphologies. In China the traditional
Hutongs feature with slum morphologies (Cat. A; I
m
of 3.81) significant
morphologic differences to urban villages (Cat. BC; I
m
of 2.57). In the
USA we find significant morphologic differences between the low-rise
infill housing in West Adams, L.A. (Cat. B, I
m
of 3.29) vs. high rise
ordered structures in the Bronx, New York City (Cat. CD; I
m
of 1.52).
However, we also observe that morphologies are reproduced in similar
manner within one country (e.g. Turkey with Gülsuyu-Gülensu in
Istanbul (Cat. AB; I
m
of 3.43), Kadifekale in Izmir (Cat. B; I
m
of 2.97) or
Karaagac-Altigagac, Ankara (Cat. B; I
m
of 2.78) or in Brazil with Turano
in Rio de Janeiro (Cat. AB; I
m
of 3.56) and Paraisópolis in Sao Paulo
(Cat. B; I
m
of 3.30)).
At the continental level Arrival Cities obviously feature a large variety
of morphologies and we basically identify almost all categories at every
continent. The category morphologic slums (Cat. A) is found typical for
cities of the Global South: America (e.g. Petare, Caracas; I
m
of 4.03),
Africa (Kibera, Nairobi; I
m
of 4.36) and Asia (e.g. Tondo, Manila; I
m
of
4.19). Categories AB, B, BC generally are found to exist at every con-
tinent.
Fig. 7 exemplifies LoD-1 building models for different categories
based on the developed morphologic index. From the high dense, low
rise slum-like structures in Kibera Nairobi (Cat. A) to the structured
high rise pattern in Évry (Cat. CD) the visualization aims to illustrate
this morphological transition.
5.4. Forms of Arrival Cities not fulfilling the spatial concept of analysis
Beyond the identified three main and three transitional categories of
building categories in Arrival Cities, we identified other physical ap-
pearances of Arrival Cities (or types of living conditions) in the litera-
ture survey. They are not part of our morphologic analysis of settle-
ments patterns due to their incongruity with the applied spatial
concept, as they consist e.g. of just one building, a very small amount of
shelters, or they have no shelters at all. However, for a more compre-
hensive perspective on the physical appearances of Arrival Cities and
respective living conditions we present these forms in a descriptive
way. Table 2 introduces four further categories (E-H) and lists examples.
6. Discussion and interpretation
When you walk through the city, one immediately senses the
neighborhood. The physical appearance of the built environment plays
a decisive role for its atmosphere –rich, poor, safe, unsafe, busy or
sleepy. One particular part in cities –the so called Arrival City –fulfills
a specific function: cheap living spaces for the poor to get access to
urban functions. The common perception of these areas is very much
connected to slums. While slums might be the most prominent example
suggesting a globally uniform morphologic type, this study reveals and
documents the large morphological variety of Arrival Cities (or places
of the urban poor) which existed or exist across the globe. We ac-
knowledge that the variability of physical manifestations of Arrival
Cities is inexhaustible, and that their morphology is often highly dy-
namic, which is not captured comprehensively in this study either.
Although we entered an age of a massive increase of (geo-)data, the
highly resolved geoinformation (LoD1) necessary for the morphologic
analysis of these places is still widely absent. A morphologic catalogue,
an inventory, or even global maps of these places are inexistent and
only very few examples have been documented in explicit, quantitative
spatial manner at this geometric level (Kuffer, Pfeffer & Sliuzas, 2016).
In consequence, taking stock of morphologic appearances con-
ceptualized as Arrival City with 44 test sites across the globe in LoD1
functions as empirical baseline to reduce these knowledge gaps and
approximate existing categories across the globe. However, as we refer
to the selection of 44 Arrival Cities it is likely that our categorization of
morphologic types using maxima and minima is biased by our samples
and the analysis might need an adjustment if other Arrival Cities may
feature more unexpected morphologic appearances.
6.1. Capabilities for mapping of Arrival Cities
Remote sensing functions here as crucial data source –as it is,
compared to e.g. census data consistent, up-to-date and available across
the globe –and it has the capability to spatially capture certain loca-
tions of the urban poor based on the assumption of morphologic cor-
relation. This study enlarges empirical knowledge on the morphological
characteristics of Arrival Cities and presents a catalogue of measured
structural categories. We acknowledge that measurement of spatial
structures and patterns is complex and calculation methods, spatial
units of analysis or thematic dimensions have significant influence
(Taubenböck, Standfuß, Klotz, & Wurm, 2016). However, we rely on
the block units suggested in literature for capturing the features of in-
dividual building structures. Beyond, we are in line with the structural
features from the ontologies presented by Kohli et al. (2012) and Sliuzas
et al. (2008). In consequence, we assume the physical appearance is
measured in a transparent and reproducible way. Using these features
we detect three main morphological categories and three respective
transitional forms as well as four categories not fitting into the pre-
sented spatial concept. This result of seven main categories and three
transitional classes reveals the morphologic complexity and variety of
such places.
In a way, this also reflects current challenges regarding conceptual
complexity and inconsistent usages of various terms in literature (slum,
informal settlement, ghettos, etc.) aiming at urban poor. This leads to
imprecise target classes and varying measurement methods. And, it
consequently is intrinsically leading to incomparable data and biased
conclusions. The morphological perspective taken here is, naturally,
only one part of a comprehensive understanding of such areas; how-
ever, with the adopted umbrella concept of the Arrival City a distinct
catalogue of morphological categories allows systematic documentation
of such places. With this approach, we are not targeting a pre-defined
class which is conceptually fuzzy, internationally not agreed, or am-
biguously defined; rather, this allows finding distinct morphological
types without restrictions by conceptual ambiguity, legal status, or
such.
The measured large morphological variety reveals that we have to
accept that solely remote sensing as single discipline is not capable for a
global classification of Arrival Cities and related urban poverty in a
complete and unambiguous way. Some forms of the measured mor-
phology do not differ sufficiently to other physical appearances within
the urban built landscape (e.g. Categories C and CD). In consequence, it
is legitimate to reduce EO-based applications to capturing urban pov-
erty to the categories ‘morphologic slums’(Cat. A) and ‘slum-like
morphologies’(Cat. AB), as these are morphologically the most sig-
nificant categories found anywhere across the globe. This is confirmed
as it has been shown that these reflect the social group of urban poor
with high correlations (e.g. Sandborn & Engstrom, 2016; Wurm &
Taubenböck, 2018). However, the categories B and BC, although con-
taining characteristic morphologic features of Arrival Cities, reflect
hybrid forms. This means they contain significant deviations from the
stated morphologic assumptions and, in addition, their relative mor-
phological differences to other, surrounding morphological appear-
ances often become more and more marginal. Naturally, the social
groups residing there are also mixed. In consequence, a comprehensive
detection of Arrival Cities, especially for the latter cases, needs
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
163

Fig. 7. Selected examples of 3-D building models of the defined categories A, AB, B, BC, C and CD.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
164

additional data sources (such as data on income, education, among
others). This shows that EO data and classification algorithms are fea-
sible to contribute significantly to a global catalogue or inventory of
Arrival Cities (categories A and AB), but the morphologic catalogue also
reveals that a comprehensive global inventory needs multi-disciplinary
data and methods.
6.2. Determinants of morphologic appearances: theory and evidence
What remains in this discussion is the question what might de-
termine the seemingly idiosyncratic physical appearances that evolved.
The most widely acknowledged causality is the natural landscape;
against the general acceptance, we found that when classifying the 44
test sites into generally flat and generally hilly terrains (using a digital
terrain model; cf. Table A-1 appendix), the structural features in each
group do not significantly differ (e.g. median building densities show
57.4% for hilly vs. 54.9% for flat terrains). If we base this comparative
analysis only on slums (Cat. A) and slum-like (Cat. AB) structures
(I
m
> 3.25), we find building densities in hilly terrain with 66.1%
slightly denser than in flat terrain with 64.8%. Admittedly, although
there are no major differences, we find that in our sample steep terrains
cause higher utilization of space.
Another defining issue of idiosyncratic appearances is pre-urban land
division (Kostof, 1991), which can be influenced by topographical issues
but goes further involving patterns of ownership (e.g. pre-existing rural
properties or current informal land developers) or the disposition of
previous usages (e.g. farming practices). Here we find that areas with
clear pre-existing geometric street patterns for the categories A, AB, and
B (we selected unambiguous examples of Imbaba, Paraisópolis, West
Adams, Eslamsahar, Nadezhda and Lo Prado) feature a median building
density of 66.8%. This is significantly lower compared to areas of no
obvious pre-existing geometric order (we selected unambiguous ex-
amples of Makkoko, Kibera, Petare, Tondo, Dharavi, Bharat Nagar and
Santosh Nagar) featuring 75.6%. The remaining Arrival Cities are a
mixture of both classes and are thus not considered. This result lets us
assume that in areas of no pre-existing order space is utilized to a higher
degree. Pre-existing patterns also lead to a more clear geometric order:
For slums (Cat. A) and slum-like (Cat. AB) structures (I
m
> 3.25) we
see building orientations for these areas with 0.065 more geometric
than without (0.108).
In general, we find the two most prominent theories –topography
and pre-existing patterns –tested here do influence urban patterns. But,
we find there is nothing completely instinctive or predestined regarding
evolving urban patterns –either planned or unplanned. Landscape
features may sometimes be embraced, but others may also be rejected.
So, it is the complexity of local determinants within globalized pro-
cesses that form these structures.
7. Conclusion and outlook
This study documents that there is not one global morphologic
settlement type solely characteristic for Arrival Cities. And we show
that remote sensing is a crucial data source for detecting and char-
acterizing built environments of the urban poor as they are still widely
neglected in official maps. This empirical baseline study can now
function as initial point to develop EO-based classification algorithms
beyond proof of concept for morphologic slum and slum-like areas.
Beyond, matching the spatial knowledge for morphologically insignif-
icant areas with location-based information of urban geography may
allow a more systematic, consistent approach of localizing patterns,
quantities and forms of poor urban living on global scale. With it we
add a step towards a comprehensive morphological catalogue or even
an inventory as foundation for urban geography and studies about
urban poverty. Or, with a development agenda perspective, this study
may provide additional knowledge for a more comprehensive regis-
tration of the dimension and distribution of urban poverty.
“We shape our buildings; thereafter they shape us”. This famous quote
of Winston Churchill reveals how built environments shape societies. In
times of the largest migration ever –from rural areas into cities (UN,
2014) with cities extensively sprawling into their hinterlands
(Taubenböck et al., 2012), informal developers have become the largest
builders of housing in the world. They shape a large share of living
environments in Arrival Cities. In consequence, the responsibility for
shaping societies by the built environment is widely left to them. This is
not necessarily negative, as many of the former European downtown
slums have turned into beloved spots of today (Glaeser, 2010). It is of
course not the built environment alone that defines whether Arrival
Cities become an integral part of the urban landscape and society, but it
is argued to play a relevant role. In light of this it is alarming that
although we are aware of this issue as well as of the dramatic global
process of migration into cities, systematic spatial knowledge on the
physical appearance of Arrival Cities is still underrepresented.
Table 2
Categories of physical appearances of Arrival Cities not being selected for the morphologic analysis.
Cat. Terminology Description Samples & literature
E Small Infill occupation Informal occupation of small urban empty spaces (e.g. by tents or
makeshift shelters; or so called ‘laneway alleys’are an often
informal way (and sometimes tolerated by city officials) for urban
density increase by housing infill)
e.g. the Curvy, Berlin, Germany (Wehner, 2015); Terras do Lelo
Martins Lissabon, Portugal (Campos Costa et al., 2013); laneway
alleys in Hamilton or Toronto, Canada (Cubitt, 2008).
F Illegal squatting in and at
existing structures
Roof top dwellers (informal top up urban densification virtueing
and replenishing space); basement suites (Informal occupation or
illegal squatting of basements); formally planned structures
(informal occupation of formal structures, e.g. when unfinished due
to construction stops); other forms of illegal squatting include
converting cubicles, verandas, staircases into living environments.
Roof top dwellers, e.g. Hong-Kong, China; Singapore, Singapore
Pomeroy (2012); basement suites, e.g. Calgary, Canada (Tanasescu,
Wing-tak, & Smart, 2010); e.g. Torre de Davide in Caracas,
Venezuela (Davis, 2007); e.g. converting cubicles verandas,
staircases into living environments in Hong Kong (Schmitt, 2007).
G Trailer homes/Traveller
camps/Mobile homes/Boat
people
A group with a nomadic life for example in caravan pitches or boats;
e.g. in Great Britain for about 25%, legal caravan pitches are
inexistent and lead to informal parking; or in Hong Kong boat
people finding homes in cargo boats, houseboats, small fishing
crafts ashore close to the city
e.g. Dale Farm, Basildon, England; Hayes road, Sully, Wales (Porter
& Taylor, 2010); or, 0.4.% of Australians are counted ‘caravan park
residents’(Greenhalgh, 2002); boat people in Hong Kong (Schmitt,
2007) or Australia (O'Doherty & Lecouteur, 2007)
H Houseless/Homeless/Roofless
population
e.g. pavement dwellers includes people sleeping on streets without
any or inadequate shelter
e.g. pavement dwellers; e.g. in Indian cities such as Calcutta,
Mumbai, or in Dhaka, Bangladesh (Padgett & Priyam, 2017)orin
Australia where 0.5 of the population is reported to be without
shelter (ABS, 2012); for more detailed classifications see (European
Federation of National Organizations Working with the Homeless,
2017)
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
165

Acknowledgements
We sincerely thank our supporters for all the dedication to this
project: P. Aravena-Pelizari, L. Banzer, S. Brandstätter, V. Färber, S.
Grund, M. Kühnl, P. Majhen, M. Neumann, I. E. Pistopoulou, I.
Standfuß, M. Weigand. Beyond we thank European Space Imaging
(EUSI) for providing high resolution optical satellite data. This work has
received also funding from the European Research Council (ERC) under
the European Union’s Horizon 2020 research and innovation pro-
gramme (grant agreement No [714087]- So2Sat).
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://dx.
doi.org/10.1016/j.apgeog.2018.02.002.
References
ABS (2012). Australian bureau of statistics. Census http://www.abs.gov.au/census.
Bähr, J., & Mertins, G. (2000). Marginalviertel in Großstädten der Dritten Welt.
Geographische Rundschau, 52,7–8 19-26.
Baud, I., Kuffer, M., Pfeffer, K., Sliuzas, R., & Karuppannan (2010). Understanding het-
erogeneity in metropolitan India: The added value of remote sensing data for ana-
lyzing sub-standard residential areas. International Journal of Applied Earth Observation
and Geoinformation, 12(5), 359–374.
Burdett, R., & Sudjic (2007). The endless city. New York: Phaidon 510 p.
Campos Costa, P., Moreira, P., & Vinagre, V. (2013). Noutra Costa da Caparica
(21.07.2017) http://www.jornalarquitectos.pt/noutra-costa-da-caparica/.
Cubitt, E. A. (2008). Laneway infill –Re-creating an urban housing typologyOntario, Canada:
University of Waterloo PhD Thesis.
Davis, M. (2007). Planet of slums. London, New York.
Duque, J. C., Patino, J. E., Ruiz, L. A., & Pardo-Pascual, J. E. (2015). Measuring intra-
urban poverty using land cover and texture metrics derived from remote sensing data.
Lanscape and Urban Planning, 135,11–21.
Engstrom, R., Sandborn, A., Yu, Q., Burgdorfer, J., Stow, D., Weeks, J., et al. (2015).
Mapping slums using spatial features. Accra, Ghana, joint urban remote sensing event
2015 Lausanne, Switzerland.
European Federation of National Organizations Working with the Homeless (2017).
ETHOS –European typology of homelessness and housing exclusion. http://www.feantsa.
org/download/en-16822651433655843804.pdf.
Ghani, E., & Ranbur, R. (2015). Urbanization and (In)formalization. In E. Glaeser, & E.
Ghani (Eds.). The urban imperative (pp. 173–209). New Delhi: Oxford University Press.
Gilbert (2007). The return of the slum: Does language matter? International Journal of
Urban and Regional Research, 31(4), 697–713.
Glaeser, E. (2010). Triumph of the city. Penguin Press HC 352 S.
Graesser, J., Cheriyadat, A., Vatsavai, R. R., Chandola, V., Long, J., & Bright, E. (2012).
Image based characterization of formal and informal neighborhoods in an urban
landscape. IEEE Journal of Selected Topics in Applied Earth Observation and Remote
Sensing, 5(4), 1164–1176 Aug. 2012.
Greenhalgh, E. (2002). Caravan park closures: Losing affordable housing or increasing
homelessness? Parity, 15(6), 18–19.
Gueguen, L. (2014). Classifying compound structures in satellite images: A compressed
representation for fast queries. IEEE Transactions on Geoscience and Remote Sensing,
53(4), 1803–1818. http://dx.doi.org/10.1109/TGRS.2014.2348864.
Hofmann, P. (2001). Detecting informal settlements from IKONOS image data using
methods of object oriented image analysis –an example from Cape Town (South
Africa). In: Proc. 2nd int. symp. remote sensing of urban areas, regensburg, Germany, Jun.
22–23, 2001 (pp. 107–118). .
Hofmann, P., Strobl, J., Blaschke, T., & Kux, H. (2008). Detecting informal settlements
from QuickBird data in Rio de Janeiro using an object based approach. In T. Blaschke,
S. Lang, & G. J. Hay (Eds.). Lecture notes in geoinformation and cartography (pp. 530–
553). Berlin, Heidelberg: Springer. http://dx.doi.org/10.1007/978-3-540-77058-
9_29.
Huchzermeyer, M., & Karam, A. (2006). The continuing challenge of informal settlements: An
introduction. Juta/UCT Press.
Jain, S. (2007). Use of IKONOS satellite data to identify informal settlements in
Dehradun, India. International Journal of Remote Sensing, 28, 3227–3233.
Kit, O., & Lüdeke, M. (2013). Automated detection of slum area change in Hyderabad,
India using multitemporal satellite imagery. ISPRS Journal of Photogrammetry and
Remote Sensing, 83, 130–137. http://dx.doi.org/10.1016/j.isprsjprs.2013.06.009.
Kit, O., Lüdeke, M., & Reckien, D. (2012). Texture-based identification of urban slums in
Hyderabad, India using remote sensing data. Applied Geography, 32(2), 660–667
March 2012.
Kit, O., Lüdeke, M., & Reckien, D. (2013). Defining the bull's eye: Satellite imagery-as-
sisted slum population assessment in hyderabad, India. Urban Geography, 34,
413–424.
Klotz, M., Wurm, M., Zhu, X., & Taubenböck, H. (2017). Digital deserts on the ground and
from space –an experimental spatial analysis combining social network and earth
observation data in megacity Mumbai. IEEE-CPS joint urban remote sensing event
(JURSE), Dubai, VAE.
Kohli, D., Stein, A., & Sliuzas, R. (2016). Urban Slum detection using texture and spatial
metrics derived from satellite imagery. Journal of Spatial Science, 61(2), 405–426.
Kohli, D., Sliuzas, R., Kerle, N., & Stein, A. (2012). An ontology of slums for image-based
classification. Computers, Environment and Urban Systems, 36(2), 154–163.
Kostof, S. (1991). The city shaped. Urban patterns and meanings through history. Thames &
Hudson 352 p.
Kuffer, M., & Barros, J. (2011). Urban morphology of unplanned settlements: The use of
spatial metrics in VHR remotely sensed images. Procedia Environmental Sciences,
7(011), 152–157.
Kuffer, M., Pfeffer, K., & Sliuzas, R. (2016a). Slums from space—15 years of slum mapping
using remote sensing. Remote Sensing, 8(6), 455–464. http://dx.doi.org/10.3390/
rs8060455.
Kuffer, M., Pfeffer, K., Sliuzas, R., & Baud, I. (2016b). Extraction of slum areas from VHR
imagery using GLCM variance. IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing, 9(5), 1830–1840. http://dx.doi.org/10.1109/
JSTARS.2016.2538563.
Kuffer, M., Pfeffer, K., Sliuzas, R., Baud, I., & Maarseveen, M. (2017). Capturing the di-
versity of deprived areas with image-based features: The case of Mumbai. Remote
Sensing, 9(4).
Niebergall, S., Loew, A., & Mauser, W. (2008). Integrative assessment of informal set-
tlements using VHR remote sensing data –the Delhi case study. IEEE Journal of
Selected Topics in Applied Earth Observations and Remote Sensing, 1(3).
Owen, K. K., & Wong, D. W. (2013). An approach to differentiate informal settlements
using spectral, texture, geomorphology and road accessibility metrics. Applied
Geography, 38, 107–118.
O'Doherty, K., & Lecouteur, A. (2007). “Asylum seekers”,“boat people”and “illegal im-
migrants”: Social categorization in the media. Australian Journal of Psycology, 59(1),
1–12.
Padgett, D. K., & Priyam, P. (2017). Pavement dwelling in Delhi, India: An ethnographic
account of survival on the margins. Human Organization: Spring, 76(1), 73–81.
Pomeroy, J. (2012). Room at the top –the roof as an alternative habitable/Social space in
the Singapore context. Journal of Urban Design, 17(3), 413–424.
Porter, M., & Taylor, B. (2010). Gypsies and travellers. In P. Thane (Ed.). Unequal britain
(pp. 71–104). London: Continuumbooks.
Sandborn, A., & Engstrom, R. N. (2016). Determining the relationship between census
data and spatial features derived from high-resolution imagery in Accra, Ghana. IEEE
Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 9(5),
1970–1977.
Satterthwaite, D. (2003). The millennium development goald and urban poverty reduc-
tion: Great expectations and nonsense statistics. Environment and Urbanization, 15(2),
179–190.
Saunders, D. (2010). Arrival cities. London, UK: Blessing.
Schmitt, R. C. (2007). Implications of density in Hong Kong. Journal of the American
Institute of Planners, 29(3), 210–217. http://dx.doi.org/10.1080/
01944366308978065.
Recovering of slums. In R. Schneider-Sliwa, & M. Bhatt (Eds.). Determinants of poverty and
upward social mobility in urban slums. Case studies from India. Basel: Schwabe/Basel
Development Studies.
Shekhar, S. (2012). Detecting slums from QuickBird data in Pune using an object oriented
approach. The International Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences, 39(8), 519–524.
Sliuzas, R., Mboup, G., & de Sherbinin, A. (2008). Report of the expert group meeting on slum
identification and mapping Report by CIESIN, UN-HABITAT, ITC. P. 36.
Stasolla, M., & Gamba, P. (2008). Spatial indexes for the extraction of formal and informal
human settlements from high-resolution SAR images. IEEE Journal of Selected Topics in
Applied Earth Observations and Remote Sensing, 1(2) June 2008.
Sudjic, D. (2010). Managed chaos. In R. Burdett, & D. Sudjic (Eds.). Living in the endless
city (pp. 86–89). New York: Phaidon.
Tanasescu, A., Wing-tak, E. C., & Smart, A. (2010). Tops and bottoms: State tolerance of
illegal housing in Hong Kong and Calgary. Habitat International, 34(4), 478–484.
Taubenböck, H., Esch, T., Felbier, A., Wiesner, M., Roth, A., & Dech, S. (2012).
Monitoring of mega cities from space. Remote Sensing of Environment, 117, 162–176.
Taubenböck, H., & Kraff, N. J. (2014). The physical face of slums: A structural comparison
of slums in Mumbai, India, based on remotely sensed data. Journal of Housing and the
Built Environment, 29(1), 15–38 [Online first 2013], print 03/2014) March 2014.
Taubenböck, H., & Kraff, N. J. (2015). Das globale Gesicht urbaner Armut?
Siedlungsstrukturen in Slums. SpringerSpektrum107–119.
Taubenböck, H., Standfuß, I., Klotz, M., & Wurm, M. (2016). The physical density of the
city –deconstruction of the delusive density measure with evidence from European
megacities. ISPRS International Journal of Geo-Information, 5(11), 1–24.
Taubenböck, H., & Wurm, M. (2015). Ich weiß, dass ich nichts weiß –
Bevölkerungsschätzung in der Megacity Mumbai. In H. Taubenböck, M. Wurm, T.
Esch, & S. Dech (Eds.). Globale Urbanisierung –perspektive aus dem All (pp. 171–178).
Springer.
Taubenböck, H., Wurm, M., Setiadi, N., Gebert, N., Roth, A., Strunz, G., et al. (2009).
Integrating Remote Sensing and Social Science –the correlation of urban morphology
with socioeconomic parameters. IEEE-CPS urban remote sensing joint event (JURSE),
Shanghai, China (pp. 7). .
Tiwari, G. (2007). Informality and its discontents. In R. Brudett, & D. Sudjic (Eds.). The
endless city (pp. 349–351). .
UN (2015). United nations millennium development goals Report 2015http://www.un.org/
millenniumgoals/news.shtml.
UN- HABITAT (2003). The challenge of slums: Global report on human settlements
2003London, UK; Sterling, VA, USA: Earthscan Publications Ltd.
UN-HABITAT (2015) Kenya: http://mirror.unhabitat.org/content.asp?typeid=19&
catid=548&cid=4962.
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
166

United Nations (2014). World urbanization prospects. New York.
UTT - Urban Think Tank (2016). Si/No: The architecture of urban-think tank. Exhibition at
Architekturmuseum TU Munich. 19.11.2015-21.02.2016.
Vaz, L. F., & Berenstein, J. P. (2004). In K. Stanilov, & B. C. Scheer (Eds.). Morphological
diversity in the squatter settlements of Rio de Janeiro, Suburban Form: An International
Perspective (pp. 61–72). New York: Routledge.
Wehner, P. (2015). Cuvry-Brache in Berlin-Kreuzberg. Stinkefinger gegen den Investor
(29.02.2016) http://www.sueddeutsche.de/panorama/cuvry-brache-in-berlin-
kreuzberg-stinkefinger-gegen-den-investor-1.2537162.
Werthmann, C., & Bridger, J. (2016). Metropolis nonformal. Applied Research + Design,
238.
World Migration Report (2015). Migrants and cities. Geneva, Switzerland: International
Organization for Migration.
Wurm, M., & Taubenböck, H. (2018). Detecting social groups from space - remote sen-
sing-based mapping of morphological slums and assessment with income data.
Remote Sensing Letters, 9(1), 41–50.
Wurm, M., Taubenböck, H., Weigand, M., & Schmitt, A. (2017a). Slum mapping based on
multi-scale texture features in polarimetric SAR data. Remote Sensing of Environment,
194, 190–204.
Wurm, M., Weigand, M., Schmitt, A., Geiß, C., & Taubenböck, H. (2017b). Exploitation of
textural and morphological image features in Sentinel-2A data for slum mapping.
IEEE-CPS joint urban remote sensing event (JURSE), Dubai, VAE (pp. 4). .
H. Taubenböck et al. Applied Geography 92 (2018) 150–167
167