Applied Geography 79 (2017) 1e10

Contents lists available at ScienceDirect

Applied Geography
journal homepage: www.elsevier.com/locate/apgeog

Frontiers of urbanization: Identifying and explaining urbanization hot

spots in the south of Mexico City using human and remote sensing
Juan Miguel Rodriguez Lopez a, *, Katharina Heider b, Jürgen Scheffran b
a
Center for Sustainable University (KNU) and CliSAP, University of Hamburg, Grindelberg 5, 20144 Hamburg, Germany
b
Research Group Climate Change and Security (CLISEC), Institute of Geography and CliSAP/ CEN, University of Hamburg, Grindelberg 5, 20144 Hamburg,
Germany

a r t i c l e i n f o a b s t r a c t

Article history: This article applies a multi-method approach to develop a better measurement of urbanization dynamics
Received 14 March 2016 using remote and human sensing based on a GIS platform. The results demonstrate the benefits of
Received in revised form bringing human and remote sensing sources together in a framework of hot spot analysis for a megacity
29 November 2016
such as Mexico City. Furthermore, our analysis suggests that human and remote sensing work well
Accepted 1 December 2016
Available online 19 December 2016
together in detecting the expansion of illegal urban settlements. Looking at the driving factors of illegal
settlements, the existence of strong association between the expansion of illegal urban settlements and
socioeconomic factors such as unemployment, provides some answers and reveals new questions. Illegal
Keywords:
Human sensing
urban growth often leads to the loss of ecological areas in the urban frontiers, especially in areas where
Remote sensing the urbanization potential is high. As a consequence, there are conflicts with legal settlers who dislike the
Urbanization illegal expansion. This approach can be extended to and replicated in new urbanizing areas, in particular
Mexico City in Africa and Asia.
Hot spots © 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
GIS (http://creativecommons.org/licenses/by/4.0/).

1. Introduction drivers of growth are compelling this illegal urban settlement

expansion? More specifically, which indicators of urban expansion
Since many cities have been growing as a result of globalization are available from remote and human data sources and how are
and will continue to expand in the coming decades, increasing ef- these related?
forts to improve planning in integrated sustainable land use are We have selected the city with one of the biggest accumulations
needed (DESA, 2014). This development is further promoted of illegal settlements in the world: Mexico City (Davis, 2006). This
through advanced communication technologies, falling trans- urban agglomeration serves as a case study with a focus on the
portation costs and the centralization of economic activities, urban/peri-urban spaces in which a horizontal expansion of illegal
combined with rising added values in cities (IPCC, 2014). Studying urban settlement has taken place (Aguilar, 2008; Aguilar & Santos,
these processes is especially challenging because this form of 2011). Our project goal is to confirm and visualize observed trends
expansion is associated with a growing trend of illegal urbanization from the social sciences, combining satellite imagery for land use
in the developing world and related social and environmental change analysis, human sensing based on Volunteered Geographic
problems (Benítez, Pe
rez-Va

zquez, Nava-Tablada, Equihua, & Information (VGI) and the relevant existing census and statistical


LuisAlvarez-Palacios, 2012). data about this urban expansion.
The key research question of this study focuses on the expansion Regarding the question on the visualization and better under-
of horizontal illegal urban settlements along the frontiers of cities: standing of estimating the illegal settlements, this study jointly
How can this form of expanding urbanization be estimated and analyses data from different acquisition contexts (remote sensing,
how accurate are these estimations? A second aspect is: Which VGI, census). The research on urban growth benefits from newly
available information in a complaints database and high definition
remote sensing imagery obtained from the RapidEye Science
* Corresponding author. Archive (RESA) Program.
E-mail addresses: miguel.rodriguez@uni-hamburg.de (J.M. Rodriguez Lopez), For this article, we take a new approach that combines human
info@katharina-heider.de (K. Heider), juergen.scheffran@uni-hamburg.de and remote sensing to assess urbanization hot spots. Remote
(J. Scheffran).

http://dx.doi.org/10.1016/j.apgeog.2016.12.001
0143-6228/© 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
2 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10

security and supply of local products (Losada et al., 1998). The

Abbreviations gentrification process downtown also additionally intensified the
requirements of land use in the peri-urban city (Salinas, 2013a;,
AGEB

Areas Geoestadísticas Ba
sicas 2013b). However, this is only one side of the discussion because this
DESA Department of Economic and Social Affairs city expansion also promoted the construction of public services
(hospitals, schools) in the peri-urban regions (Me
ndez-Lemus,
DiB Data in Brief
GIS Geographic Information System 2012).
HS Human sensing This article is organized into six sections. After this introduction,
INEGI Instituto Nacional de Estadística y Geografía Section 2 outlines the theory. In Section 3, the analysis of the data
IPCC Intergovernmental Panel on Climate Change with the proposed methods is presented. Sections 4 and 5 contain
PAOT Procuraduria Ambiental y del Ordenamiento results and discussion and Section 6 summarizes the work and
Territorial del Distrito Federal draws conclusions.
RESA RapidEye Science Archive
RS Remote sensing 2. Theory and approach
VGI Volunteered Geographic Information
This article presents a multi-method approach to a better
measure of urbanization dynamics. Our study investigates the im-
plications of rapid urban expansion for land competition and the
sensing is based on satellite information and is one of the most environment at the frontier of the city. The methods and findings
prominent and developed sources of geographical information for are of direct practical relevance, given that urban populations are
land use research. The data gathering method of human sensing is projected to double to 6.4 billion by 2050, and the issues related to
related to an emerging field of literature, which is using the po- urbanization will increase in importance (IPCC, 2014), as will
tentials of the Web 2.0 to create new and reliable databases for related conflicts and environmental problems that need to be
geographical research (Haklay, Basiouka, Antoniou, & Ather, 2010). resolved in a sustainable manner. Besides its direct relevance for
In both fields, the question of better measurements is a key issue. In Mexico City, the ultimate goal of our research is to develop a
this study, remote sensing detects urban change with high- methodological approach that is appropriate for direct transfer and
resolution satellite imagery (5 m) from the RapidEye sensor, diffusion to other cities and regions, specifically to the newly
while human sensing detects illegal settlements in the form of flourishing cities in Africa and Asia, which have seen similar pat-
complaints in a georeferenced point data set. This article addresses terns of growth, but have not reached the urbanization rate of
the question to which degree the estimations of human and remote Mexico.
sensing are comparable and complementary, exploring whether Our goal is to confirm and visualize observed trends from the
and how the benefits of one technique could compensate for the social sciences using human sensing (VGI) and remote sensing
deficits of the other. (satellite imagery), combining it with the relevant existing census
In Mexico City, it is not possible for the administration to avoid and statistical data regarding illegal settlements in Mexico City. VGI
unplanned urban development and as a consequence, an increasing has become popular with help of new Web 2.0 technologies;
number of nature reserves and related ecosystem services such as however, the question of quality concerning this kind of data re-
water regeneration and CO2 storage (Delgado-Ramos, 2015; mains open (Bimonte, Boucelma, Machabert, & Sellami, 2014). The
Delgado-Ramos, Mancheno, & Rodriguez Lopez, 2014) become policies of Internet companies have allowed the use of this infor-
lost in unplanned settlement processes. Even existing regulations mation (e.g. API Twitter) and Google Earth has become a respect-
have so far failed to contain the expansion (Aguilar, 2008; Aguilar & able tool for mapping with high-resolution images and at different
Santos, 2011). The lack of planning and defined land use has led to points in time (Clark & Aide, 2011). Furthermore, Copernicus or
tensions in Mexico City between the city's center and the periphery. Landsat programs provide the opportunity to obtain satellite in-
In the city center, land prices are so high that the majority of the formation of different qualities for almost all parts of the world and
population settles in the cheaper suburbs. Streams of people travel at various times.
in the morning into the city center for work and return in the The uses of VGI vary from emergency responses and urbaniza-
evening to the periphery (Lomelí, 2008). This phenomenon of peri- tion decisions to a valuable tool in data science (Elwood, 2010).
urbanization has, as a consequence, produced a growing demand However, research about VGI is needed to prove how effective it is
for infrastructure, particularly transport and services such as water and under which circumstances it works best (Goodchild &
and energy supply, but also for accommodation, so that people stay Glennon, 2010).
overnight in the peri-urban settlements. Various forms of remote sensing data for rural-urban in-
One of the most well-known historical cases is the district teractions are available for the vast majority of countries. We pro-
Nezahualcoyotl in the metropolitan area of Mexico City (Linares pose to complement these data with human sensing and use the
Zarco, 2013). In addition to undeveloped areas and nature re- opportunity here to match various forms of data with remote
serves being irregular and illegal settlements occupied by the sensing imagery to generate new insights (Hagenlocher, Lang, &
poorer population (Platt, 2010), real estate speculations are also Tiede, 2012). By combining satellite imagery and VGI in a
transforming the area (Aguilar, 2008; Aguilar & Santos, 2011). The Geographic Information System (GIS) platform, we aim, for
latter are implicitly promoted with the support or permissive example, to differentiate between settled and un-settled land, as
attitude of the authorities (Aguilar & Santos, 2011). A particular well as used resources (water, food and energy), over the peri-
problem is that the expansive urbanization with illegal settlements urban areas. Furthermore, census and statistical or qualitative
is occupying an area of “great ecological value in terms of climate data gathering are expensive and burdensome means of collecting
regulation, water recharge, forest communities, agricultural culti- information (McCann, Colby, Easter, Kasterine, & Kuperan, 2005).
vation, and hilly landscape” (Aguilar, 2008, p. 133). As a result these As the cost of collecting satellite imagery continues to fall and VGI
areas are losing worth (Jujnovsky, Gonza
lez-Martínez, Cantoral- forms of data continue to develop, huge potential exists in
Uriza, & Almeida-Len ~ ero, 2012), with the steady loss of food combining remote sensing with human sensing data to generate
increased value for governments, NGOs, and researchers. This will
 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10 3

complement and enhance the understanding of the reasons behind Mexico City, allows analysis of the relationship between the illegal
the expansion of illegal settlements, providing a powerful means of and legal settlers in more detail, including the linkages between
visualizing a complex social phenomenon. poverty and the loss of ecological values (Aguilar & Guerrero, 2013).
As central topics, this project addresses the information The first assumption is that informal settlers generally need new
retrieval and analysis of VGI and the social context, especially territory, which is possible to find in the South, but is detrimental to
regarding better ways of measuring the information and the rela- the natural reserve (Jujnovsky et al., 2012).
tion between subjective classification, like in the case of remote The technical discussion about remote sensing provides many
sensing, and the common ontology of the cartography. Combining examples of efforts to map illegal settlements. These attempts try to
human and remote sensing, this article brings two streams of capture the development of illegal settlements by identifying the
literature together. In order to understand more about the better proper shape and structure of the illegal settlement (Hacker et al.,
measuring of both databases, hot spot analysis was carried out to 2013). Most of these papers work with a combination of images,
bring the information to scale. To illustrate this analysis, several and they apply a segmentation analysis (Başkent & Kadiog ullari
figures accompany this article. 2007; Kit & Lüdeke, 2013; Taubenbo €ck & Kraff 2014). This
Part of the literature discusses the best architecture for forming method produces acceptable results in creating a classification of
new data with VGI. There are two different approaches; on the one illegal settlement forms (Kohli, Sliuzas, Kerle, & Stein, 2012), but it
hand, there is a system with data collectors or custodians, while on requires a high density of housing typical for a consolidated illegal
the other hand, the second approach argues for a crowdsourcing settlement (Kit, Lüdeke, & Reckien, 2012). Alone, this view of
and free forming of the data. Most of the systems today fall between remote sensing is often not enough and in low populated rural and
both approaches and this discussion remains open (Kalantari, semi-rural areas, the algorithms based on density and fragmenta-
Rajabifard, Olfat, & Williamson, 2014). The wiki-social-movement tion have limitations. This method works much better in slums in
could be observed as an example, as this tool works relatively the urban areas (Ward und Peters 2007). The more applied research
well with the rules and authorship in WikiGenes (Hoffmann, 2008). demonstrates the power of combining land use change and socio-
More geographically, attempts with OpenStreetMap have shown economic surveys (Wilebore und Coomes 2016).
promising assessments in environmental planning of land use Moreover, a broader view should include not only the expansion
(Jokar Arsanjani, Helbich, Bakillah, Hagenauer, & Zipf, 2013). of settlements, but should also address this topic of illegality. The
However, VGI is still a debatable field because quality cannot be dimension of the problem shows the need for an integrating
easily guaranteed and the answer to the question of how to approach to face the challenge of urbanization in the frontiers of
convince people to work for the generation of information is not cities (Marx, Stoker, & Suri, 2013). We assume here that poor people
clear (Fritz et al., 2009). Since maps are not a perfect representation are stigmatized and criminalized more for their acts when they
of space, this project is conscientious about cartography as a affect the interests of the richer classes. The plausibility of this
continuously evolving tool, and how cartography is changing argument is high and researchers have been analysing this hy-
already existing maps (Kitchin & Dodge, 2007). A prominent pothesis in other contexts (see Gustafson, 2009 for the US
example concerning research on Wikipedia shows that where the example). From a societal critical perspective, we argue that public
number of users working on one article increases significantly, the policies (like the conservation area in Mexico City) allow for this
quality of information increases as well, when we compare these conflict of interest between the illegal settlers and the legal settlers
articles with peer review articles of the Encyclopaedia Britannica (Gustafson, 2011). Therefore, hypothesis 2 for the research question
(see Giles, 2005). A similar effect can be observed for the geography is that socioeconomic factors positively influence settlement
with OpenStreetMap. When the number of users of OpenStreetMap expansion in the periphery of the city: the higher the unemployment
working on one map increases significantly, the quality of infor- rate, the lower the level of identification of illegal settlements.
mation increases as well (Haklay et al., 2010). This concept is known
as “Linus’ Law”. Therefore, we argue for a similar positive rela- 3. Data and methods
tionship between the number of inhabitants and the overlapping of
the VGI and remote sensing information. Mexico City's rural-urban area is located inland, in the middle of
In this study the VGI information consists of complaints about the southern part of the country (see Fig. 1). This growing urban
illegal settlements, and we compare it with remote sensing of ur- space has placed a high demand on resources. Since Mexico City
ban expansion data, leading to hypothesis 1: the more populated an depends on the water flowing from outside the urban space (e.g.
area is, the higher the overlap of VGI and remote sensing data. the Magdalena River south of Mexico City), active water manage-
Parts of the expansion of illegal settlements have been well ment is needed to maintain the water supply and the ecosystem,
researched in detail (Aguilar, 2008; Aguilar & Santos, 2011). How- especially in the southern area, called the “preservation zone,”
ever, a holistic, simultaneous consideration of all these spatial data within the nature reserve (Jujnovsky et al., 2012). Therefore, the
in relation to socioeconomic data and VGI is still missing. Therefore, study area falls in the Federal District of Mexico City and the focus
this article investigates the connection between socioeconomic lies particularly on this preservation zone. To analyse this area, a
variables, remote sensing analysis and VGI across various spatial grid with a cell size adapted to the study area was created. On the
scales of this complex dynamic system, assuming that the maps are one hand, using the grid is a good mean to overcome stiff man-
not a perfect representation of space. The definition of illegality in made borders. On the other hand, it is necessary for our analysis
Mexico City is a good example of how the political situation is because the administrative areas in the conservation zone are too
already changing existing maps (Kitchin & Dodge, 2007). Part of rough (Rodriguez Lopez, Heider, & Scheffran, 2017).
this process is the construction of categories like ‘legal and illegal An analysis of locations with high urbanization concentration
settlements’ in the context of continuous growth in the South and has been conducted using two different data sets based on remote
the historical steady legalization of occupied territory (Aguilar & and human sensing (see Fig. 2 for a visualization of the data). Using
Santos, 2011; Aguilar, 2008; Jujnovsky et al., 2012). This analysis Geographic Information Systems (GIS) makes it possible to esti-
contributes to explaining the relationship between land use and mate urbanization hot spots in the southern part of Mexico City
conflict regarding illegal and legal settlements. from both data perspectives, utilizing the commercial software
Above all, the possibility that quantitative and qualitative ArcGIS 10.3.
treatment are integrated by crossing remote and human sensing in The remote sensing data is based on high-resolution imagery
4 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10

Fig. 1. Location map and the study area in the southern part of Mexico City (Research area in red).

Fig. 2. Location of data in the southern part of the Federal District in Mexico City: urban change, calculated from 5 m remote sensing imagery from the RESA program, and ecological
complaints (human sensing) from PAOT.

(5 m) of the RapidEye satellite classification from a seminal 2017). Two classes (urban and non-urban) sum up the final set of
research project in Mexico City (Rodriguez Lopez, Rosso, Scheffran, land cover classes, considering that the main goal of this study is to
& Delgado Ramos, 2015), and its technical description is docu- analyse the impact of urbanization on protected areas. Our main
mented in the journal Data in Brief (DiB) (Rodriguez Lopez et al., focus is the mapping of urban and forested areas because the latter
 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10 5

are often associated with protected areas. In the end, the change
detection class of areas that were non-urban in 2009 and urban in
2014 is extracted and in the following analysis is compared with a
second data set based on the complaints filed at the “Procuraduria
Ambiental y del Ordenamiento Territorial del Distrito Federal”
(PAOT).1 Fig. 2 displays both databases in the research area.
Altogether the information in the PAOT dataset includes almost
18,000 georeferenced complaints between 2002 and 2013. For a
detailed discussion of the complaints, see the supplemental DiB
paper (Rodriguez Lopez et al., 2017). For the analysis only com-
plaints about illegal land use are included in a time frame between
2002 and 2013. Thus, the VGI data start before the first satellite
image was taken which is due to the possibility that people need
time to get accustomed to the possibility of reporting illegal set-
tlements. There can be persons that report a complaint immedi-
ately and others who report it after years. The key assumption for
our analysis is that the urban expansion should produce an increase
in the number of complaints regarding illegal settlements within
the research area.
The data analysis is divided into four main steps. In the first two
steps hot spots of urbanization and complaints are identified in two
different ways. On the one hand, hot spots are detected by identi-
fying high values of ecological complaints (human sensing) and
urbanization (remote sensing) densities. On the other hand, an
optimized hot spot analysis is implemented, using the Getis-Ord
Gi* statistic for ecological complaints and urbanization detected
by remote sensing imagery. Using an optimized hot spot analysis
helps to deal with the quality issues of VGI. In an optimized hot spot
analysis a higher weight is given to spatial clustering which further
strengthens observations of many people for the same or nearby
locations (Goodchild & Li, 2012). In the next step, the hot spot
analysis combines the results and detects overlaps between the
data sets of urbanization and complaints. Finally socioeconomic
drivers are analysed and integrated to find a relationship between
population density, unemployment rate and urbanization, as well
Fig. 3. Overview of methods, combining human and remote sensing.
as complaint hot spots (see Fig. 3). We chose population density
since one of the most popular arguments in the literature about the
efficiency of VGI argues with the number of individuals who are identification of hot and cold spots. For a more detailed description
taking part in the generation of the database. One typical socio- of the method, see the supplemental paper about the data
economic feature is unemployment, which is easy to replicate in (Rodriguez Lopez et al., 2017).
other databases and was inserted from the used National Census To calculate how much of the urbanization observed by remote
2010 database (National Census Bureau in Mexico, INEGI, 2010). sensing is explained by human sensing, the number of cells
The unemployment rate is used as a good estimator of the socio- belonging to the hot spots is analysed (see Tables 1 and 2). This
economic level of the population, as established in one area. Al- analysis is conducted for the main study area (Federal District) with
ternatives include poverty or income, which are numbers further the cell size used in the main analysis (561 m) as well as for a higher
discussed in this context. resolution (378 m).
In the following we merge and compare spatial statistics and For the integration of socioeconomic drivers, population and
remote sensing classification methods. The spatial statistics unemployment data of Mexico City, the smallest available basic
approach is known as cluster analysis, which is a good opportunity geostatistical area (AGEB), is drawn from the National Census 2010
to bring both databases into the same scale, observing the spatial (National Census Bureau in Mexico, INEGI, 2010). Due to the high
distribution of illegal settlements and the associated conflicts be- resolution of the geostatistical area, a higher resolution (378 m)
tween legal and illegal settlers. In this analysis, each pixel classified was chosen for this part of the analysis.2 In the next step, the data is
as non-urban in 2009 and urban in 2014 is converted into one classified in four quantiles and the number of cells within a hot spot
point. To identify hot spots directly, the top quantile of the number belonging to the corresponding class is identified (Tables 3 and 4). If
of complaints and urbanization in a grid cell is identified in both the number of cells in one class exceeds one quarter (of the total
data sets. Additionally, the optimized hot spot analysis is conducted number of cells in the hot spot), the hot spot is overrepresented in
to include neighbouring cells using ArcGIS. Within the analysis, that class.3 Additionally, a t-test compares the difference in the
statistically significant hot and cold spots are identified using the
Getis-Ord Gi* statistic (Getis & Ord, 1992; 1995). In the end, the
analyses are carried out with a resolution of 561 m since it gener-
2
ates the best results for the autocorrelation test for the automatic The available basic geostatistical areas (AGEB) have a high variety of different
sizes. While the smallest area is 0.006 km2, the biggest area is 7.25 km2. The mean
size is 0.4 km2 with a standard deviation of 0.504 km2. To minimize loss of accuracy
and stay consistent within our analysis, we chose a grid of 0.143 km2.
1 3
Governmental office in charge of reviewing and processing environmental and We thank one of the reviewers for this idea of analysis in overrepresentation of
land use offences. quantiles.
6 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10

Table 1
Percentage of overlapping between complaints (HS) and urban expansion (RS) cells with incident values in top quantile (Q 0.75).

Districts of Mexico City (cell size) HS: no. of cells in top quantile Q 0.75 RS: no. of cells in top quantile Q 0.75 Overlap of HS and RS [%]

Federal District (561 m) 62 478 12.97

Federal District (378 m) 27 923 2.92

Table 2
Percentage of overlapping between complaints (HS) and urban expansion (RS) hot spots (99% confidence) calculated with Getis-Ord GI* statistic

Districts of Mexico City (cell size) HS: no. of cells in hot spots RS: no. of cells in hot spots Overlap of HS and RS [%]

Federal District (561 m) 251 798 31.45

Federal District (378 m) 462 1636 28.24

Table 3 data sets, remote and human sensing. The databases are at the
Distribution of cells of the complaints hot spot (HS), urbanization hot spot (RS) and same level of aggregation within a fishnet representation that di-
overlap (99% confidence) in quantiles of the population density in the Federal Dis-
trict (378 m).
vides the research area into cells with a spatial resolution of 561 m
of length and width. In the following maps, the hot spots of the top
Quantiles of population density Overlap HS RS quantile as well as hot and cold spots with different confidence
Q0 117a 188a 412a levels are illustrated. The confidence level ranges from 90 to 99%.
Q 0.25 115 190a 408 The darker the blue for cold spots and the darker the red for hot
Q 0.5 116a 186 408
spots, the higher the confidence level is.
Q 0.75 114 186 408
Total no. of cells 462 750 1636 The discussion about the spatial resolution of the grid requires
a
an explanation. There are two ways of selecting the cell measure-
More than 1/4th in a quantile.
ment. The first is with an optimal value of spatial autocorrelation,
e.g. with the method of incremental spatial autocorrelation, and the
Table 4
second takes a given value as the basis. This research is based on an
Distribution of complaints hot spot (HS), urbanization hot spot (RS) and overlap (99% optimal value of autocorrelation obtained by using the tool of
confidence) in quantiles of the unemployment rate in the Federal District (378 m). optimized hot spot analysis in ArcGIS 10.3 that identifies statisti-
Quantiles unemployment rate Overlap HS RS
cally significant spatial clusters of hot and cold spots, by aggre-
gating, scaling and testing incident data for spatial dependence.
Q0 122a 191a 426a
The first observation is that the southern edge of the remote
Q 0.25 130a 203a 442a
Q 0.5 109 185 404 sensing hot spot is located particularly near to the border of the
Q 0.75 101 171 364 preservation zone. The hot spot analysis of land use change related
Total no. of cells 462 750 1636 to city expansion with remote sensing (Fig. 4) shows a more
a
More than 1/4th in a quantile. extensive cover of the research area than the hot spot analysis of
human complaints (Fig. 5). This was expected since not every
change from non-urban to urban land use implicates an increment
means of hot spots and non-hot spots for the two variables (see of the illegal settlements or a complaint. Nevertheless, it was ex-
Table 5). pected that the complaint hot spots would fit within the land use
change category regarding the change from non-urban to urban
4. Results land use, in other words, the horizontal expansion of the city.
The results of the overlapping human and remote sensing data
Using the methodology described, urbanization and complaint can be seen in Figs. 6 and 7. In Fig. 6, the individual layers of the
hot spots are identified in two different ways. The direct analysis human sensing and the remote sensing hot spots are displayed in a
identifies cells in the top quantile concerning the number of com- more transparent lighter colour, while the dark red indicates the
plaints or urbanization pixels per cell. The optimized hot spot hot spot overlap with 99% confidence of both layers. This is the
analysis using Getis-Ord Gi* identifies hot and cold spots in both central figure for the first question of the article, concerning how

Table 5
t-test results comparing hot spots and non-hot spots of population density and unemployment rate.

Variables Min Max Mean Std. Dev. No. of cells t-test p (t) value

Population density [sqkm] HS Overlap 0 20,106 6091 3261 462 0.0001 (11.62)
NHS Overlap 0 49,397 2958 5716 5501
HS Remote Sensing 0 31,508 8091 5375 1636 0.0001 (48.82)
NHS Remote Sensing 0 49,397 1352 4500 4327
HS Complaints 0 20,106 4554 3500 750 0.0001 (7.07)
NHS Complaints 0 49,397 3006 5845 5213
Unemployment rate [%] HS Overlap 0 10.2 4.325 1.833 462 0.0001 (25.79)
NHS Overlap 0 25 1.464a 2.324 5501
HS Remote Sensing 0 10.2 4.163 1.82 1636 0.0001 (62.89)
NHS Remote Sensing 0 25 0.748a 1.89 4327
HS Complaints 0 10.2 3.675 2.318 750 0.0001 (25.42)
NHS Complaints 0 25 1.399a 2.289 5213
a
The low values of the unemployment rate are caused by the dark figure in the preservation zone.
 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10 7

Fig. 4. Optimized hot spot analysis of city expansion in the Federal District: 5 m remote sensing imagery from the RESA program aggregated to 561 m fishnet resolution.

Fig. 5. Optimized hot spot analysis of city expansion in the Federal District with human sensing: ecological complaints from PAOT aggregated to 561 m fishnet resolution.

accurate the estimations of urban expansion are, in this case, by urban expansion. The cells on which the table is based contain
human and remote sensing. In Fig. 7, the direct analysis uses the incident values for remotely detected urbanization or complaints.
number of complaints and urbanization cases per grid cell in the The top quantile (75 percent of the distribution are below this
top quantile to identify hot spots that show a similar location of the quantile) of incident values per grid cell was analysed for cell sizes
overlap as in Fig. 6. The dark red area indicates the overlap of the of 561 m and 378 m. Combining both analyses, we calculated how
hot spot while the light red illustrates the hot spot based either on much of the observed urbanization by remote sensing overlapped
human or remote sensing. Due to the exclusion of neighbouring with human sensing. 12.97% of urbanization detected by remote
cells, the hot spots, and consequently, the overlaps, are smaller. sensing is explained by complaints from human sensing. For the
Tables 1 and 2 summarize the results of the overlap between smaller resolution of 378 m, 2.92% of remote sensing can be
human and remote sensing. In Table 1 the percentage of overlap for explained by human sensing. In the same way, we compared the
the two databases used in the project with densities in the top results of the optimized hot spot analysis. The percentage of over-
quantile are shown. The human sensing data represents the lap of highly significant hot spots for the two databases used in the
ecological complaints and the remote sensing data illustrates the project with different cell sizes is illustrated in Table 2. Human
8 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10

Fig. 6. Results of optimized hot spot analysis of data overlaid from remote sensing (5 m resolution from RESA program) and human sensing (ecological complaints from PAOT).

Fig. 7. Results of hot spot analysis of data in the top quantile overlaid from remote sensing (5 m resolution from RESA program) and human sensing (ecological complaints from
PAOT).

sensing overlaps with 31.45% (at 99% confidence) of the observed Getis-Ord Gi* statistic. Table 3 indicates the distribution of the ur-
urbanization identified by remote sensing (at 99% confidence) for a ban expansion, the ecological complaints, as well as the over-
resolution of 561 m and 28.24% for a resolution of 378 m (see lapping hot spots by different population densities in the research
Table 2). The overlapping results of optimized hot spots and the hot area. The distribution of the various forms of hot spots through
spots calculated from a direct comparison of density values differ population density helps to test hypothesis 1. The population
from each other because the optimized hot spot analysis in- densities are classified in four quantiles. If the number of cells in
corporates neighbouring cells in a hot spot where incidents are one class exceeds one quarter, the hot spot is overrepresented in
present. These incidents might be below the top quantile in the this class. Table 3 shows that the hot spots are lightly over-
direct analysis of incidents per cell in Table 1. With increasing represented in the classes with low population densities.
spatial resolution of cell size, the incidents can be located more For testing hypothesis 2, Table 4 presents the distribution of the
precisely which leads to less overlap. urban expansion, the ecological complaints, as well as the over-
In Tables 3 and 4 the correlation with socioeconomic factors is lapping hot spots by different unemployment rates in the research
analysed. The analysis is based on the hot spots calculated with the area. The unemployment rate is classified in four quantiles as
 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10 9

described for Table 3. The hot spots of remote sensing, human detected by satellites although the complaints data set
sensing and the overlap are clearly overrepresented in classes with (2002e2013) starts before the first satellite imagery was taken
low unemployment rates, as can be seen in Table 4. (2009), possibly because urbanization follows the complaints. Thus,
Table 5 shows the results and necessary data for the t-test. Hot in the interaction between urbanization and complaints, timing
spots and non-hot spots are tested for their difference in means. matters, in particular, whether the complaints occur before or after
This is conducted for both variables. A statistically significant dif- the urbanization is detected by satellite images. To analyse this,
ference between hot spots and non-hot spots is detected. better temporal resolution is required.
A limitation of this article is the definition of trustworthiness as
5. Discussion of the results “better measurement”, which should be more clearly specified and
analysed. A second limitation is found in the fact that this article
In Tables 1 and 2 the effect of using different cell sizes can be looks at one city and one time frame. Further research is needed,
seen. While the best result of 31.45% overlap is reached with the and particular attention should be paid to areas where the potential
roughest resolution of 561 m, the percentage of overlap between of urbanization is high in Mexico City and in other regions, espe-
the complaints hot spot and the urbanization hot spot decreases cially in the growing cities of Asia and Africa.
with a higher resolution of 378 m. The results for the optimized hot
spot analysis which includes neighbouring cells (Table 2) delivers 6. Conclusions
better results than the direct analysis of incident values in the top
quantile (Table 1). Remote and human sensing have been presented as important
After the identification of hot spots (Figs. 4e7), we checked for new sources of information, but several practical questions arise
the difference in means between hot spot and non-hot spot areas. surrounding the measurement of these sources. We proposed a
In Table 5 the results of the t-test are illustrated. In all cases there is multi-method approach to improve measuring the urbanization
a significant difference between hot spots and non-hot spots. This is dynamics and the use of a combined remote and human sensing
not a surprising result because within the non-hot spot lies most of analysis.
the conservation zone, which is not included in the AGEB data and The results indicate that it is possible to bring two databases
is here considered with a 0 for no population and no unemploy- from human and remote sensing sources together with a hot spot
ment. This causes very low unemployment rates and is clearly a analysis in a megacity region. Furthermore, our findings suggest
problem but it is no alternative to exclude the conservation zone that VGI works well in detecting the expansion of illegal urban
because it is an important area for the identification of complaints settlements, and associated complaints among settlers. The exis-
and urbanization hot spots, especially in the border area. Another tence of a strong association between the expansion of illegal urban
problem that should be discussed as a consequence of the lack of settlements and socioeconomic factors opens new possibilities of
AGEB data in the conservation zone is the spatial autocorrelation in research to discover more about this kind of urbanization. There-
the population and socioeconomic data. In both data sets, spatial fore, further research into the causes of urbanization and its asso-
autocorrelation can be measured using the Moran's I test. This ciated conflicts is desirable.
might affect the robustness of the results from this part of the This proposed method could be readily used in the growing
analysis and has to be addressed in future research. cities of Africa and Asia, however there is a limitation concerning
In the next step, the correlation of hot spots with demographic the existence of human sensing information in the form of geore-
and socioeconomic variables is analysed. The relationship is not ferenced data on conflicts across different scales. Remote sensing
what is expected. Observing the results of the quantile analysis data is available for a large number of countries and years, and
regarding the population density and the identification of hot spots although it is not always free, access to this data has become easier.
in Table 3, it can be seen that the hot spots in the lower quantiles of This research has great potential for the application of new forms of
low population densities are lightly overrepresented. Therefore, we “big data” such as microblogging (e.g. Twitter) or other georefer-
can reject the first hypothesis that the more populated an area is, enced databases (e.g. Openstreetmaps). Nevertheless, it requires
the greater the number of identified hot spots and the measure- relatively specialized programming skills in georeferencing and
ment of the VGI. We observe an overrepresentation of cells but in database compilation.
the other direction, the reasons for which are possibly in the so- It is fundamental for policy makers to understand the process of
cioeconomic aspects of the urban expansion. Alternative reasons to illegal urbanization, in order to manage this urban expansion and to
explain this unexpected result could be based on migration, eco- form mitigation and adaptation strategies in the context of climate
nomic growth, housing market policies, infrastructure, and local change (IPCC, 2014). The challenge is to identify the forces behind
regulation (Roy, Lees, Palavalli, Pfeffer, & Sloot, 2014). More urban growth, analyse the drivers of the settlement expansion and
research is needed in this area. analyse the impact of this dynamics that may lead to the loss of
The chosen socioeconomic variable (unemployment rate) shows ecological areas in the urban periphery, especially in areas where
a strong negative association with the identification of human and the urbanization potential is high. In these places, low-carbon ur-
remote sensing hot spots, as expected in hypothesis 2 (see Table 4). banization could be an important mitigation strategy. Assuming
All hot spots are overrepresented in the classes with low unem- monetary compensation for mitigation, this could create an op-
ployment rates. The total of the complaints’ hot spots works portunity for the people living there and empower them with an
especially well in correlation with the unemployment numbers. To asset, or it could be a risk for them because of their diffuse property
understand these results, some clues can be found in Fig. 2, which rights. In any case, it will be a possibility and a risk for illegal set-
shows the locations of the complaints that are largely located on tlements to make use of their rights for a life of dignity.
the urban fringe, or edge of the urbanization periphery, where the
greatest changes occur. Only in a few cases complaints occur within Acknowledgments
urban boundaries. This indicates that complaints are strongest in
the areas affected by new urbanization, which also have a low This work was supported in parts by Centre for a Sustainable
population density and at the same time an increasing employment University (KNU - Generation 3) and by the Cluster of Excellence
rate. It is interesting to look at the complaints on the pathways to “Integrated Climate System Analysis and Prediction” (CliSAP -
the south and southeast, which are not matched by urbanization EXC177) funded through the German Science Foundation (DFG).
10 J.M. Rodriguez Lopez et al. / Applied Geography 79 (2017) 1e10

Underlying RapidEye data has been contributed on behalf of the Hoffmann, R. (2008). A wiki for the life sciences where authorship matters. Nature
genetics, 40(9), 1047e1051. http://dx.doi.org/10.1038/ng.f.217.
German Aerospace Center through funding by the German Federal
INEGI. (2010e2012). Estadísticas Censales a Escalas Geoelectorales. Principales resul-
Ministry of Economy and Energy. The authors would like to thank tados del Censo de Poblacio
n y Vivienda. http://gaia.inegi.org.mx/geoelectoral/
Andrea Balbo for useful comments and discussions on an early doctos/FD_SECC_IFE.pdf (Accessed 17 December 2015).
version of the paper, as well as the proofreading of John Elflein and IPCC. (2014). Climate change 2014. Mitigation of climate change. Cambridge, United
Kingdom: Cambridge University Press.
student assistance of Johanna Remus. Jokar Arsanjani, J., Helbich, M., Bakillah, M., Hagenauer, J., & Zipf, A. (2013). Toward
mapping land-use patterns from volunteered geographic information. Interna-
References tional Journal of Geographical Information Science, 27(12), 2264e2278. http://
dx.doi.org/10.1080/13658816.2013.800871.
Jujnovsky, J., Gonza
lez-Martínez, T. M., Cantoral-Uriza, E. A., & Almeida-Len ~ ero, L.
Aguilar, A. G. (2008). Peri-urbanization, illegal settlements and environmental (2012). Assessment of water supply as an ecosystem service in a rural-urban
impact in Mexico City. Cities, 25(3), 133e145. http://dx.doi.org/10.1016/
watershed in southwestern Mexico City. Environmental Management, 49(3),
j.cities.2008.02.003. 690e702. http://dx.doi.org/10.1007/s00267-011-9804-3.
Aguilar, A. G., & Guerrero, F. L. (2013). Poverty in Peripheral Informal Settlements in Kalantari, M., Rajabifard, A., Olfat, H., & Williamson, I. (2014). Geospatial metadata
Mexico City: The Case of Magdalena Contreras, Federal District. Journal of 2.0ean approach for volunteered geographic information. Computers. Envi-
Economic and Social Geography, 104(3), 359e378. http://dx.doi.org/10.1111/ ronment and Urban Systems, 48, 35e48. http://dx.doi.org/10.1016/
tesg.12012.
j.compenvurbsys.2014.06.005.
Aguilar, A. G., & Santos, C. (2011). Informal settlements' needs and environmental Kitchin, R., & Dodge, M. (2007). Rethinking maps. Progress in Human Geography,
conservation in Mexico City: An unsolved challenge for land-use policy. Land
31(3), 331e344. http://dx.doi.org/10.1177/0309132507077082.
Use Policy, 28(4), 649e662. http://dx.doi.org/10.1016/j.landusepol.2010.11.002. Kit, O., & Lüdeke, M. (2013). Automated detection of slum area change in Hyder-
ullari, A. I. (2007). Spatial and temporal dynamics of land use
Başkent, E. Z., & Kadiog
_ abad, India using multitemporal satellite imagery. ISPRS Journal of Photogram-
pattern in Turkey. A case study in Ineg €l. Landscape and Urban Planning, 81(4),
o
metry and Remote Sensing, 83, 130e137. http://dx.doi.org/10.1016/
316e327. http://dx.doi.org/10.1016/j.landurbplan.2007.01.007. j.isprsjprs.2013.06.009.
Benítez, G., Pe
rez-V


azquez, A., Nava-Tablada, M., Equihua, M., & Luis Alvarez-
Kit, O., Lüdeke, M., & Reckien, D. (2012). Texture-based identification of urban slums
Palacios, J. (2012). Urban expansion and the environmental effects of informal in Hyderabad, India using remote sensing data. Applied Geography, 32(2),
settlements on the outskirts of Xalapa city, Veracruz, Mexico. Environment and 660e667. http://dx.doi.org/10.1016/j.apgeog.2011.07.016.
Urbanization, 24(1), 149e166. http://dx.doi.org/10.1177/0956247812437520. Kohli, D., Sliuzas, R., Kerle, N., & Stein, A. (2012). An ontology of slums for image-
Bimonte, S., Boucelma, O., Machabert, O., & Sellami, S. (2014). A new spatial OLAP based classification. Computers. Environment and Urban Systems, 36(2),
approach for the analysis of volunteered geographic information. Computers,
154e163. http://dx.doi.org/10.1016/j.compenvurbsys.2011.11.001.
Environment and Urban Systems, 48, 111e123. http://dx.doi.org/10.1016/ Linares Zarco, J. (2013). Nezahualco
yotl, de ciudad dormitorio a polo de desarrollo
j.compenvurbsys.2014.07.006. de la regio
n Oriente del Valle de Me
xico. Paradigma Econo
mico, 5(2), 117e144.
Clark, M. L., & Aide, T. M. (2011). Virtual interpretation of Earth Web-Interface Tool Lomelí, L. (2008). In R. Cordera, P. R. Kuri, & A. Ziccardi (Eds.), Pobreza, desigualdad y
(VIEW-IT) for collecting land-use/land-cover reference data. Remote Sensing, exclusio
n social en la ciudad del siglo XXI. Me
xico: Siglo XXI.
3(3), 601e620. http://dx.doi.org/10.3390/rs3030601.
s, J. (1998). Urban
Losada, H., Martinez, H., Vieyra, J., Pealing, R., Zavala, R., & Corte
Davis, M. (2006). Planet of slums. New York: Verso. agriculture in the metropolitan zone of Mexico city: Changes over time in ur-
Delgado-Ramos, G. C. (2015). Water and the political ecology of urban metabolism:
ban, suburban and peri-urban areas. Environment and Urbanization, 10(2),
The case of Mexico city. Journal of Political Ecology, 22, 98e114. 37e54. http://dx.doi.org/10.1177/095624789801000214.
Delgado-Ramos, G. C., Mancheno, T., & Rodriguez Lopez, J. M. (2014). New latin Marx, B., Stoker, T., & Suri, T. (2013). The economics of slums in the developing
american perspectives on sustainable and low carbon societies. Global Trans- world. Journal of Economic Perspectives, 27(4), 187e210. http://dx.doi.org/
formations towards a Low Carbon Society. Working Article Series, 10. 10.1257/jep.27.4.187.
DESA. (2014). World Urbanization Prospects: The 2014 Revision. New York: United
McCann, L., Colby, B., Easter, K. W., Kasterine, A., & Kuperan, K. V. (2005). Trans-
Nations: Department of Economic and Social Affairs. action cost measurement for evaluating environmental policies. Ecological
Elwood, S. (2010). Geographic information science: Emerging research on the so- Economics, 52(4), 527e542. http://dx.doi.org/10.1016/j.ecolecon.2004.08.002.
cietal implications of the geospatial web. Progress in Human Geography, 34(3), Me
ndez-Lemus, Y. (2012). Urban growth and transformation of the livelihoods of
349e357. http://dx.doi.org/10.1177/0309132509340711. poor campesino households: The difficulties of making a living in the periphery
Fritz, S., McCallum, I., Schill, C., Perger, C., Grillmayer, R., Achard, F., et al. (2009).
of Mexico city. International Development Planning Review, 34(4), 409e438.
Geo-Wiki. Org: The use of crowdsourcing to improve global land cover. Remote http://dx.doi.org/10.3828/idpr.2012.25.
Sensing, 1(3), 345e354. http://dx.doi.org/10.3390/rs1030345.
Platt, H. L. (2010). Exploding cities: Housing the masses in Paris, Chicago, and
Getis, A., & Ord, J. K. (1992). The analysis of spatial association by use of distance Mexico city, 1850d2000. Journal of Urban History, 36(5), 575e593. http://
statistics. Geographical Analysis, 24(3), 189e206. http://dx.doi.org/10.1111/ dx.doi.org/10.1177/0096144210365454.
j.1538-4632.1992.tb00261.x. Rodriguez Lopez, J. M., Heider, K., & Scheffran, J. (2017). Human and remote sensing
Getis, A., & Ord, J. K. (1995). Local spatial autocorrelation Statistic: Distributional data to investigate the frontiers of urbanization in the south of Mexico city. Data in
issues and an application. Geographical Analysis, 27(4), 286e306. http://
Brief, submitted.
dx.doi.org/10.1111/j.1538-4632.1995.tb00912.x. Rodriguez Lopez, J. M., Rosso, P., Scheffran, J., & Delgado Ramos, G. C. (2015). Remote
Giles, J. (2005). Internet encyclopaedias go head to head. Nature, 438(7070), sensing of sustainable rural-urban land use in Mexico city: A qualitative anal-
900e901. http://dx.doi.org/10.1038/438900a. ysis for reliability and validity. Interdisciplina, 3(7), 321e342.
Goodchild, M. F., & Glennon, J. A. (2010). Crowdsourcing geographic information for Roy, D., Lees, M. H., Palavalli, B., Pfeffer, K., & Sloot, M. P. (2014). The emergence of
disaster response: A research frontier. International Journal of Digital Earth, 3(3),
slums: A contemporary view on simulation models. Environmental Modelling &
231e241. http://dx.doi.org/10.1080/17538941003759255. Software, 59, 76e90. http://dx.doi.org/10.1016/j.envsoft.2014.05.004.
Goodchild, M. F., & Li, L. (2012). Assuring the quality of volunteered geographic
n en la ciudad latinoamericana. El caso de Buenos
Salinas, L. (2013a). Gentrificacio
information. Spatial Statistics, 1, 110e120. http://dx.doi.org/10.1016/ Aires y Ciudad de Me
xico. Geographos. Revista digital para estudiantes de Geo-
j.spasta.2012.03.002. grafía y Ciencias Sociales, 44(4), 281e304. http://dx.doi.org/10.14198/
Gustafson, K. S. (2009). Criminal law. The criminalization of poverty. The Journal of GEOGRA2013.4.44.
criminal law & criminology, 643, 668e669. Salinas, L. (2013b). La gentrificacio
n de la colonia Condesa, Ciudad de Me
xico.
Gustafson, K. S. (2011). Cheating welfare. Public assistance and the criminalization of
n desde Latinoame
rica. Revista Geogra
fica de Am

Aporte para una discusio erica
poverty. New York University Press. Central, 2(51), 145e167.
Hacker, K. P., Seto, K. C., Costa, F., Corburn, J., Reis, M. G., Ko, A. I., et al. (2013). Urban Taubenbo € ck, H., & Kraff, N. J. (2014). The physical face of slums. A structural com-
slum structure. Integrating socioeconomic and land cover data to model slum parison of slums in Mumbai, India, based on remotely sensed data. Journal of
evolution in Salvador, Brazil. International Journal of Health Geographics, 12, 45. Housing and the Built Environment, 29(1), 15e38. http://dx.doi.org/10.1007/
http://dx.doi.org/10.1186/1476-072X-12-45.
s10901-013-9333-x.
Hagenlocher, M., Lang, S., & Tiede, D. (2012). Integrated assessment of the envi- Ward, P. M., & Peters, P. A. (2007). Self-help housing and informal homesteading in
ronmental impact of an IDP camp in Sudan based on very high resolution multi-
peri-urban America. Settlement identification using digital imagery and GIS.
temporal satellite imagery. Remote Sensing of Environment, 126, 27e38. http:// Habitat International, 31(2), 205e218. http://dx.doi.org/10.1016/
dx.doi.org/10.1016/j.rse.2012.08.010. j.habitatint.2007.02.001.
Haklay, M., Basiouka, S., Antoniou, V., & Ather, A. (2010). How many volunteers does
Wilebore, B., & Coomes, D. (2016). Combining spatial data with survey data im-
it take to map an area well? The validity of linus' law to volunteered geographic proves predictions of boundaries between settlements. Applied Geography, 77,
information. The Cartographic Journal, 47(4), 315e322. http://dx.doi.org/
1e7. http://dx.doi.org/10.1016/j.apgeog.2016.09.007.
10.1179/000870410X12911304958827.