ArticlePDF Available

Missing "Middle Scenarios": Uncovering Nuanced Conditions in Latin America's Housing Crisis

Authors:

Abstract

This article proposes a novel approach to capturing housing deficiencies in rapidly urbanizing regions that is more suitable for data capture, policymaking, and redevelopment. Housing deficit data need to be accurately captured and categorized to adequately act on them. As underscored in the New Urban Agenda, urban policymakers need to be able to set accurate and realizable targets to address the housing crisis. Local governments require precise knowledge of ground realities to strategically allocate money and resources, target solutions, and avoid needless waste. Planners and architects require detailed data and plans to respond to complex conditions. Citizens and nongovernmen-tal organizations should be informed of their communities' needs to engage and collaborate in enhancing local urban service delivery. The goal of this article is threefold: (1) to explore a new method to rapidly capture high-quality housing data in the region; (2) to discuss how these deficits and spatial patterns could be clustered into a new form of place-based deficiency typologies; and (3) to contribute to a more precise analysis of housing shortfalls for planners, policymakers, and local governments using the Latin American region as a case-study scenario. The proposed solution has the potential to be lower cost, more accurate, rapid, and scal-able compared with currently applied techniques and technologies such as census data and satellite imagery. This research serves as a call to action and an exploration in the potential to leverage unmanned aircraft systems and a leading artificial intelligence machine learning algorithm applied to its output data to test, program, and study geospatial data for the purpose of capturing and categorizing rapidly changing qualitative housing typologies in Latin America.
31
Cityscape: A Journal of Policy Development and Research • Volume 19 Number 2 • 2017
U.S. Department of Housing and Urban Development • Office of Policy Development and Research
Cityscape
Missing ‘Middle Scenarios’:
Uncovering Nuanced Conditions
in Latin America’s Housing Crisis
Kira Intrator
Massachusetts Institute of Technology
Kaustubh Shivdikar
Veermata Jijabai Technological Institute
Abstract
This article proposes a novel approach to capturing housing deficiencies in rapidly
urbanizing regions that is more suitable for data capture, policymaking, and redevelop-
ment. Housing deficit data need to be accurately captured and categorized to adequately
act on them. As underscored in the New Urban Agenda, urban policymakers need to
be able to set accurate and realizable targets to address the housing crisis. Local govern-
ments require precise knowledge of ground realities to strategically allocate money and
resources, target solutions, and avoid needless waste. Planners and architects require
detailed data and plans to respond to complex conditions. Citizens and nongovernmen-
tal organizations should be informed of their communities’ needs to engage and col-
laborate in enhancing local urban service delivery. The goal of this article is threefold:
(1) to explore a new method to rapidly capture high-quality housing data in the region;
(2) to discuss how these deficits and spatial patterns could be clustered into a new form
of place-based deficiency typologies; and (3) to contribute to a more precise analysis of
housing shortfalls for planners, policymakers, and local governments using the Latin
American region as a case-study scenario.
The proposed solution has the potential to be lower cost, more accurate, rapid, and scal-
able compared with currently applied techniques and technologies such as census data
and satellite imagery. This research serves as a call to action and an exploration in the
potential to leverage unmanned aircraft systems and a leading artificial intelligence ma-
chine learning algorithm applied to its output data to test, program, and study geospatial
data for the purpose of capturing and categorizing rapidly changing qualitative housing
typologies in Latin America.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
32
Intrator and Shivdikar
Habitat III
Introduction
Latin America and the Caribbean are plagued by massive housing deficits. Currently, more than 59
million individuals—one in three families—live in deficient housing that lacks basic infrastructure,
is unsuitable to inhabit, or built with low-quality materials (Bouillon, 2012). This dilemma is
compounded by the fact that Latin America is one of the fastest-growing and most-urbanized de-
veloping regions in the world (Bredenoord, van Lindert, and Smets, 2014). Formal housing has not
been able to keep pace; illegal or informal settlements have remained one of the more common,
and in some regions the primary, means of providing housing for low-income households. In addi-
tion, housing occupies the most sizeable surface area in cities. Although figures vary considerably,
the principal land use in all cities is residential, occupying from 65 to 75 percent of the surface
(Rodrigue, 2013). At the urban scale, tackling the housing challenge directly addresses the most
predominant land use in cities. However, deficits in Latin America and the Caribbean are calcu-
lated and collapsed into a simplistic binary of qualitative and quantitative shortfalls. An absence
of shared definitions, inconsistent methodologies, and outdated information compound this issue.
Although directionally helpful, the binary definition of qualitative and quantitative shortfalls is
unable to accurately capture complex urban realities. Binary definitions are static and do not help
pinpoint acute needs. Moreover, binary definitions do not measure the magnitude of specific char-
acteristics of the housing crisis at a neighborhood, urban, or regional scale. Categorizing housing
deficiencies through a binary lens helps produce binary responses. Nuanced and targeted solutions
that address ground realities are likely neglected.
Among the main pillars examined and problematized at the 2016 United Nations Conference on
Housing and Sustainable Urban Development (Habitat III) were guidelines and improvements to
national urban policy and housing policies (clustered under the key areas of urban frameworks,
and urban housing and basic services respectfully). Habitat III also addressed the need for ac-
curate and timely data capture to ensure closer alignment between local needs and national urban
planning frameworks. As stated in Habitat III’s New Urban Agenda, articles 111 and 156 affirm the
importance of capturing “differentiated analysis of housing supply … based on high-quality, timely,
and reliable disaggregated data at the national, sub-national, and local levels” and encourage the
use of digital tools such as “geospatial information systems … to improve long-term integrated
urban and territorial planning and design” (United Nations, 2017).
Drivers for Change: Why Is a New Approach Needed?
The current method used to define and analyze the housing deficit in Latin America is binary
and, therefore, limited. The housing shortage in Latin America is neither a monolithic statistic nor
binary in nature. Regional housing deficits are calculated and collapsed into a simplistic binary of
qualitative and quantitative shortfalls. Qualitative shortfalls represent the inadequacy of housing
units with respect to structural shortfalls, such as poor housing materials and quality; insufficient
space; privation of basic infrastructure (for example, electricity, sanitary disposal of waste, and po-
table water); poor location; and insecure tenure. Quantitative deficiencies denote a lack of shelter,
severe overcrowding, or a need for a completely new dwelling. Although directionally helpful, the
binary definition is unable to accurately capture complex urban realities. The definitions are static
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Missing ‘Middle Scenarios’: Uncovering Nuanced Conditions in Latin America’s Housing Crisis
33
Cityscape
and do not pinpoint acute needs, nor do they measure the magnitude of specific characteristics
of the housing crisis at the neighborhood, urban, or regional scale. Examining and categorizing
housing deficiency through a binary lens helps produce binary responses. Nuanced, targeted, and
effective solutions that address the ground reality likely are overlooked or neglected. As under-
scored throughout the New Urban Agenda, the development of robust and accurate urban policies
is vital to providing the necessary framework and support to urban development, acting as a key
instrument to capture and target current and future urban priorities and to coordinate and guide
the actions of national and local actors, as well as private and public investments.
Housing deficiencies do not exist in a binary bubble. The urban reality is much more complex,
with a wide variety of conditions that constitute “middle scenarios.” This article argues that these
missing middle scenarios—which cut across regional, urban, and neighborhood scales—need
to be explored and exposed. Imagine an agglomeration of poorly ventilated slum housing in an
urban center with weak roofs and collapsing walls surrounded by highrise buildings. Compare this
scenario with decrepit lowrise social housing in a periurban region rebuilt with deteriorating walls
and roofs and with little to no access to public transportation or schools. Although both display
similar measures of qualitative shortfalls (for example, weak walls and roofs), these two scenarios
have very different sets of conditions and solutions. An analysis of housing shortfalls more suitable
for redevelopment is desperately needed.
To generate accurate policies, urban plans, and architectural or landscape designs, middle scenario
typologies are needed to reflect the complexity of the housing challenge and condition. Although
a significant range of complex urban conditions exists, patterns of typologies likely will emerge.
These patterns need to be captured, learned, and analyzed. This article explores a solution to this
challenge; it proposes that a catalog of local, regional, and national housing deficiency typologies
be developed through pattern-recognition and machine learning on the back end of an unmanned
aerial vehicle (UAV). The goal of cataloging these deficits is to generate a new method of clustering
typologies and a more nuanced analysis of the housing deficit. In sum, the proposed solution
would help address the shortcoming of current measurements and provide a novel approach to
collect and parse housing deficiency data.
Institutional Challenges
In Latin America, an absence of shared definitions and a prevalence of inconsistent methodologies
and outdated information characterize the quantitative and qualitative shortfalls regarding housing
data. Little agreement exists in the region as to what defines qualitative deficiencies and adequate
shelter or as to the methodology of how to capture these shortfalls. Definitions vary by country and
by researcher in the field of housing, from setting a minimum predefined livable space—wherein
households or individuals living beneath this minimum requirement are classified as inadequate—
to assessing housing conditions by measuring levels of deprivation (Rojas, 2015).
There is a paucity of data. The Economic Commission for Latin America and the Caribbean
completed the most recent Latin American regional assessment in 2000, primarily by drawing on
census data from the 1990s (UN-Habitat, 2011). Census data largely exclude informal areas and
settlements. Ad hoc or independent efforts have been launched to address this data gap by the
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
34
Intrator and Shivdikar
Habitat III
Center for Social Research at the nonprofit TECHO or through an interim housing census launched
in 2015 in Brazil and Mexico (to be carried out every decade).1 Nevertheless, there remains a lack
of timely, high-quality data capture and analysis. Specifically, local and national data on informal
settlements tend to be imprecise and fragmented (Fernandes, 2011). Moreover, despite the fact that
numerous regional and international agencies attempt to provide reliable statistics and encourage
standardization and updated methodologies to measure deficiencies, a desperate need remains for
the production of standardized and periodic data.2 Census data are captured every decade. Region-
ally, the average amount of time dating to the most recent census is more than 6 1/2 years (see
exhibit 1). Outdated, inaccurate, and binary information needs to be replaced, and the absence of
shared definitions and inconsistent methodologies needs to be countered.
Exhibit 1
Census Data Captured by Year, Latin America
Notes: At the time of this research, Guatemala’s Instituto Nacional de Estadistca (INE) website was disabled and not open to
the public. All countries that conduct the housing census collect qualitative and quantitative housing data. In 2015, Mexico
and Brazil introduced an interim housing census that will be carried out every decade. Chile will create a new one-off census
in 2017 because the data captured in 2012 did not adhere to the required data standards.
Sources: Map by Kira Intrator and Ricardo Martínez Campos using data from Instituto Nacional de Estadística y Censos
(INEC; Argentina); Instituto Nacional de Estadística de Bolivia (Bolivia); Instituto Brasileiro de Geografia e Estadística (Brazil);
Instituto Nacional de Estadística (INE; Chile); Departamento Administrativo Nacional de Estadística (Colombia); INEC (Costa
Rica); Oficina Nacional de Estadística (Dominican Republic); INEC (Ecuador); Dirección General de Estadística y Censos (El
Salvador); INE (Guatemala); INE (Honduras); Instituto Nacional de Geografía, Estadística e Informática (Mexico); INEC (Nicara-
gua); INEC (Panama); Dirección General de Estadísticas, Encuestas y Censos (Paraguay); Instituto Nacional de Estadística e
Informática (Peru); INE (Uruguay); INE (Venezuela)
1 To counter the lack of data on informal settlements, the Center for Social Research conducted alternative census studies in
Argentina, Chile, Colombia, and Nicaragua in 2015.
2 Examples of regional and international agencies seeking standardized definitions, updated housing indicators,
methodologies, and regional coordination around indicators include the Regional Office on Latin America and the
Caribbean, the Latin American and Caribbean Demographic Centre, and the Global Housing Policy Indicators.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Missing ‘Middle Scenarios’: Uncovering Nuanced Conditions in Latin America’s Housing Crisis
35
Cityscape
General Scope and Scale of Challenge
The regional, urban, and neighborhood-scale impact of Latin America’s housing crisis is massive.
Latin America’s housing deficit and the concomitant rise of central and peripheral informal housing
settlements has heightened large-scale urbanism challenges. The dynamics that determine polycen-
tric or periurban expansion are complex—ranging from housing policies, sociospatial issues, a lack
of serviceable land, and weak urban planning and infrastructure—a chief result of which is the
unavailability of suitable, affordable housing.
Urban areas in Latin America suffer primarily from qualitative housing deficiencies. Of these quali-
tative deficiencies, the most pervasive is inadequate infrastructure, followed by a shortfall in the
quality of the materials used for shelter. Basic infrastructure provision includes sanitary disposal
of waste and access to electricity and potable water. Additional qualitative shortfalls include over-
crowding; poor-quality or impermanent materials; inadequate roof, wall, or floor conditions; and
tenure. Rojas and Medellin (2011) corroborated that the region’s greatest unmet need is qualitative
(exhibit 2). In addition, statistics published at a major continental ministerial meeting indicate
that 60 percent of households have access to suitable housing in the region. Of the remaining 40
percent, 22 percent live in dwellings that are structurally deficient and the final 18 percent need
entirely new houses (MINURVI, 2007). According to more recent studies in the category of infra-
structure, the leading challenge is inadequate sanitation, affecting 13 percent of urban households
(about 17 million households), and 6 percent (or 8 million households) lack piped water. Electric-
ity access appears to be inconsequential, with nearly universal coverage (Bouillon, 2012).
Exhibit 2
Urban Housing Deficiencies by Type, Latin America
Percent of total households
Quantitative Infrastructure Materials / Overcrowding
Quality
Tenure
1995 2009
35
30
25
20
15
10
5
0
Source: Adapted from Rojas and Medellin (2011)
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
36
Intrator and Shivdikar
Habitat III
Uncovering ‘Missing Middle’ Scenarios in Latin America’s
Housing Crisis
A UAV, commonly known as a drone, is a pilotless aircraft. UAVs are a component of unmanned
aircraft systems (UAS), which include a UAV, a ground-based controller, and a system of communi-
cations between the two. The flight of UAVs may operate with various degrees of autonomy, either
remotely by a human operator or fully or intermittently autonomously by onboard computers
(ICAO, 2011). UAS provide a cheaper and faster alternative to conventional mapping techniques.
The advantage of using these systems is demonstrated in the speed and accuracy with which the
provide geospatial data and their capacity to generate accurate data of existing formal or informal
spatial developments. UAS are a logical means of collecting and parsing spatial patterns and data
due to the fast pace of Latin America’s urban growth, the presence of undocumented informal
housing settlements, and the fact that housing census data are collected once every decade. The
proposed solution is for the UAS data to feed into a new form of image-based machine learning,
developed from current urban conditions. Machine learning, using the long short-term memory
(LSTM) algorithm, would help surface patterns and provide the primary input to the formulation
of these missing middle scenarios and urban spatial typologies.
UAS counter the cost, complexity, and time of conventional mapping. Multilateral agencies,
researchers, the private sector, and governments are testing the nascent field of UAS mapping. Dur-
ing the past year, for example, The World Bank used UAS to generate geospatial data for surveying
land in countries with weak or outdated geospatial data infrastructure (World Bank, 2016). At
a smaller scale, independent research initiatives experiment with UAS as a primary input for
participatory mapping and to collect geospatial data for informal housing settlements in unmapped
regions (Barnes, Volkmann, and Muller, 2015). Through seed funding from initiatives such as
Drones for Good from the United Arab Emirates, private-sector startups use UAS to undertake
a range of activities such as three-dimensional mapping, ocean navigation and surveillance, and
emergency surveillance in remote geographies (UAE Drones for Good, 2017).
In Latin America, governments primarily use UAS for activities related to intelligence, surveillance,
and reconnaissance operations, with a few governments beginning to apply alternative purposes
such as monitoring environmental crimes in the Brazilian Amazon rainforest and architectural
digs in Peru (Cawley, 2014, 2013; Emol, 2013). However, these data collection approaches remain
binary in nature and do not employ or combine artificial intelligence with UAS. The proposed
approach and solution has yet to be to be applied to defining or addressing the housing crisis in
developing regions.
The UAS could be used in a new way—to test, program, and study geospatial data for the purpose
of capturing and categorizing qualitative housing typologies in Latin America. UAS have yet to be
deployed in Latin America for the specific use of cataloging housing and their surrounding urban
spatial characteristics. This idea responds to a critical need for urban and housing deficit data that
are accurate and rapidly generated. The proposed solution could use and expand on current data
libraries as a baseline input (for example, census data, local survey data, infrastructure provision,
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Missing ‘Middle Scenarios’: Uncovering Nuanced Conditions in Latin America’s Housing Crisis
37
Cityscape
and so on) related to qualitative urban housing conditions. Moreover, the solution aims to capture
the types of shortages compromising an individual home, as well as examine the larger urban
fabric in which it resides.
The proposed concept explores a novel approach to the data mining process by using the latest
LSTM model. Predictive models, such as deep neural networks, have the potential to generate
a real-time heat map of city characteristics and housing deficiency needs. Many of the recurrent
neural network algorithms, such as the LSTM model that contains prolific nodes, have made an
impact in real-life applications. LSTM is extensively used and adapted through algorithms involv-
ing speech recognition and are applied in the field of robotics for vision analysis, scene labeling,
and face recognition. For example, in their work of speech recognition, Sak, Senior, and Beaufays
(2014) dissected acoustic frames that were analyzed using a phoneme database with the LSTM
model. Unlike the simple feed-forward neural networks, LSTM models are tantamount to cyclic
connection-based modeling sequences; through the use of LSTM, these sequences are converged
quickly, and the data are processed in real time. The proposed model is unique because of the
parameters that are input into the algorithm. Although a simple LSTM network, this model will
be peculiar in that the data provided by the drone will be processed in real time using its array of
various sensors (see exhibits 3 and 4).
Exhibit 3
Input and Output Samples for the Long Short-Term Memory Algorithm
Camera feed
Thermal imaging
Temperature data
LiDAR
Census
Correlaon
Infrastructure
water electricity
School
Transportaon
Roof and walls
Venlaon
Input Process Output
LiDAR = Light Detection and Ranging.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
38
Intrator and Shivdikar
Habitat III
Exhibit 4
Snapshot of Long Short-Term Memory Algorithm (or Reoccurring Neural Network)
Functions
Camera feed
Thermal
imaging
Temperature
data
LiDAR
Census
Infrastructure
water electricity
School
Transportaon
Roof and walls
Venlaon
Input Process Output
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Node
Latin American
Census
LiDAR = Light Detection and Ranging.
Implementing the Model
LSTM models are well known for finding both closely related and sparsely related dependencies.
This important characteristic of LSTM models can be directly correlated to our problem statement.
The short-term dependencies will develop logic nodes for medium-sized databases with immediate
connection to the output—for example, connecting light detection and ranging (LIDAR) data
to calculate the distance between objects and gather information on the number of floors of a
building. In contrast, the long-term dependencies of the model will be capable of relating abstract
concepts to one another.3
The input to the proposed solution includes UAS equipped with camera feed, thermal imaging,
and LIDAR sensors (exhibit 3). Existing census and temperature data would further be fed into the
LSTM system as external data sources. The camera feed from the UAS is of higher quality, more
detailed, less expensive, and more timely than census data collection and has the potential to be
much more scalable than existing satellite or Landsat imagery. Thermal imaging enables estimates
of how many people live in a given house or area, and LIDAR would provide basic input such
as the topography of the region and the height of buildings (for example, LIDAR combined with
thermal imaging data could be cross-referenced to calculate how many individuals live in a given
dwelling and space).
3 For example, a long-term dependency could include connecting temperature data from a region with the availability of
public transportation. Even though a heat map of a particular area may seem superfluous in identifying the availability
of public transport, such data help determine the relationship by linking the heat map data to previous data samples. If a
similar pattern of heat map variations with respect to time were found after the introduction of public transport in another
region, the algorithm would conclude the presence of public transport in an area with similar heat-map-to-time data patterns.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Missing ‘Middle Scenarios’: Uncovering Nuanced Conditions in Latin America’s Housing Crisis
39
Cityscape
A prototype of the LSTM model could be conducted in a small preselected urban agglomeration
in which input parameters are collected and the LSTM model trained through data tagging of
the region. In its maiden flight, the drone gathers data and streams it to a local database. The
LSTM model would use this data to correlate and drive conclusions about the spatial patterns
and clusters of housing deficiencies of that area. Previous census data would then be compared
with these findings and the aberrations used to fine-tune the LSTM model. Once such a network
is established, it would use the correlation (also known as learned neural network nodes) on the
surrounding houses in the area (see exhibit 4). The UAS, when deployed, will transmit massive
amounts of data to online servers for large-scale computations. An important feature of the LSTM
model is the ability to process data in real time. The model is capable of processing gathered intel-
ligence and correlating it with previous data, thus providing output in real time. Real-time output
makes the data processing faster. The algorithm would be deployed on rented servers, as the server
environment would enable easy and, in most cases, automatic expansion of the algorithm through
the self-replication of the code. This ability to extrapolate its computational capability would not
only enable multiple drone deployments but would keep the downtime of the model to a mini-
mum, due to the fact that servers have multiple redundancies. Such a model of running algorithms
on low-cost, dedicated rental servers is advantageous, as it would allow for building on the readily
available high-tech infrastructure of premium server services such as Amazon Web Services and
Google Cloud Platform. With respect to the financials of deploying the UAS, the initial hardware
costs incurred for the drone, LIDAR sensors, and thermal imagers would be substantial. After this
initial investment, longer-term recurring charges would be minimal and include only hardware
maintenance and server infrastructure charges. Moreover, the pool of big data that this type of
initiative would generate would be massively helpful to multiple sectors across industries. For
example, the real estate industry would be able to locate regions requiring redevelopment, the food
and supply industries would be able to better allocate resources, and governments would be able
to map and create requisite policies to abate high levels of pollution.
To generate correlations and categorize spatial housing deficiency patterns, the algorithm receives
data from the UAS. The algorithm will initially generate nodes, or so-called virtual decisionmakers,
and start interconnecting the nodes based on the type of data collected in real time from the drone.
This interconnection of nodes will enable users to conclude connections and map out high-quality
housing deficiency patterns through the use of remotely related parameters. For example, UAS
send data of a select region to the servers. The algorithm uses the primary data from the camera
feed to process telltale signs that certain objects are present on roofs by comparing them with its
database of objects. For example, the UAS discovers the existence of large ventilation shafts and
air-conditioning units on a roof, and the algorithm concludes that the building is well-ventilated.
Object recognition and image annotation will be carried out by matching object features, including
speeded-up robust features (SURF) and Fast Retina Keypoint (FREAK) features. The SURF and
FREAK features enable the LSTM algorithms to match corner points and edges of objects in the
camera feed. This feature comparison makes it possible for the model to detect ventilation, the
availability of water supply by detecting water pipes, and structural integrity of the buildings
through the detection of crevices in buildings and discoloration of walls. Other data, such as the
LIDAR sensor feeds, enable a topographical analysis of the region and a calculation of the height
of given infrastructure and buildings. Another critical feature of the neural network is its ability
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
40
Intrator and Shivdikar
Habitat III
to predict certain outputs in the absence of direct inputs; it extrapolates the data it has retrieved
to predict parameters it has never been fed. It does so by extrapolating from indirect inputs. For
example, if no image or sensor data of the interiors of houses exist to provide direct information
regarding levels of overcrowding or the quality of floors or ceilings, it uses thermal images and
LIDAR data to relate them to floor and ceiling conditions. Although the accuracy of the model
reduces when relying on indirect parameters, it is an added advantage nevertheless.
A limitation of the proposed model is that the sensors used by the UAS are susceptible to weather
changes. One way to tackle this issue is by adding or subtracting Delta errors to the sensor read-
ings. However, this adjustment would eliminate the error only if these weather-induced errors
remained constant when data are fed to the network. In contrast, weather induced errors fluctuate
considerably when covering a large area. This limitation would be addressed as the network grows;
with a substantial amount of data, the error is reduced by taking the simple average of repetitive
data. An additional implementation challenge to be addressed would be military areas and no-fly
zones. To address this issue, the data channeled through the network could be preprocessed to
avoid legal and privacy issues. Such an attempt can be made by selectively muting onboard sensors
when a drone flies in those areas. Many Latin American countries have their own drone regula-
tions. Panama, for example, divides drones into four categories based on weight (AAC, 2016).
Argentina requires drones to stay out of a 5-kilometer radius around airports (ANAC, 2015).
Conclusion
Housing deficit data need to be captured and categorized accurately for policymakers, urban plan-
ners, architects, citizens, and nongovernmental organizations to adequately act on it. This article
outlined a novel approach to capturing and recategorizing missing middle scenarios of rapidly
changing qualitative housing typologies in Latin America by leveraging UAVs and neural network
machine learning. The proposed solution is more suitable for data capture, policymaking, and
redevelopment in rapidly urbanizing regions and has the potential to be lower cost, more accurate,
rapid, and scalable compared with currently applied techniques and technologies such as census
data and satellite imagery.
Authors
Kira Intrator is the principal of the Civic Strategy Group and is a global partnership manager and
urban planning researcher at the Massachusetts Institute of Technology, School of Architecture and
Planning.
Kaustubh Shivdikar is an electrical engineer in the Department of Electrical Engineering at Veer-
mata Jijabai Technological Institute.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Missing ‘Middle Scenarios’: Uncovering Nuanced Conditions in Latin America’s Housing Crisis
41
Cityscape
References
Administración Nacional de Aviación Civil (ANAC). 2015. “Proyecto Normativa Para Vehículos
Aéreos No Tripulados (UAV).” http://www.anac.gov.ar/anac/web/index.php/1/1088/raacdnar-bajo-
proceso-de-elaboracion-participativa-de-normas-decreto-117203/-ltpre-gtproyecto-normativa-
para-vehiculos-ltbr-gtaereos-no-tripulados-uav-ltpre-gt.
Autoridad Aeronautica Civil (AAC). 2016. “Resolucion No. 120/DSA/DJ/DG/AAC.” http://www.
aeronautica.gob.pa/seguridad_aerea_nacional/index.php?c=seguridad_aerea_drone.
Barnes, Grenville, Walter Volkmann, and Anna Muller. 2015. “Drones in Support of Upgrading
Informal Settlements: Potential Application in Namibia.” Paper presented at 2015 World Bank
Conference on Land and Poverty, Washington, D.C., March 23–27.
Bouillon, César, ed. 2012. Room for Development: Housing Markets in Latin America and the Carib-
bean. New York: Palgrave Macmillan.
Bredenoord, Jan, Paul van Lindert, and Peer Smets. 2014. “Introduction: Governance, Sustain-
ability and Affordability of Low-Income Housing.” In Affordable Housing in the Urban Global South:
Seeking Sustainable Solutions, edited by Jan Bredenoord, Paul van Lindert, and Peer Smets. New
York: Routledge: 1–14.
Cawley, Marguerite. 2014. “Drone Use in Latin America: Dangers and Opportunities,” InSight
Crime: Investigation and Analysis of Organized Crime, April 18. http://www.insightcrime.org/news-
analysis/drone-use-in-latin-america-dangers-and-opportunities.
———. 2013. “Rising Eco-Trafficking Threatens Brazil Amazon,” InSight Crime: Investigation
and Analysis of Organized Crime, November 15. http://www.insightcrime.org/news-briefs/eco-
trafficking-threatens-brazil-amazon.
Emol. 2013. “Peruanos Amplian Uso de ‘Drones’ en Arqueologia y Agricultura,” Emol.com, August
10. http://www.emol.com/noticias/internacional/2013/08/10/613929/drones-surcan-cielos-de-
peru-para-usos-en-arqueologia-y-agricultura.html?utm_source=twitterfeed&utm_medium=twitter.
Fernandes, Edesio. 2011. Regularization of Informal Settlements in Latin America. Cambridge, MA:
Lincoln Institute of Land Policy.
General Assembly of Ministries and High Level Authorities on Housing and Urbanization in Latin
America and the Caribbean (MINURVI). 2007. “Declaration of Santiago de Chile.” Paper presented
at MINURVI XVI, Santiago, Chile, October 8–9.
International Civil Aviation Organization (ICAO). 2011. Unmanned Aircraft Systems (UAS), CIR328
AN/190. Quebec, Canada: International Civil Aviation Organization. http://www.icao.int/Meetings/
UAS/Documents/Circular 328_en.pdf.
Rodrigue, Jean-Paul. 2013. The Geography of Transport Systems, 3rd ed. New York: Routledge.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
42
Intrator and Shivdikar
Habitat III
Rojas, Eduardo. 2015. “Housing Policies, Quality of Housing and Urban Development: Lessons
From the Latin American Experience 1960–2010.” Paper presented at 2015 World Bank Confer-
ence on Land and Poverty, Washington, D.C., March 23–27.
Rojas, Eduardo, and Nadine Medellín. 2011. Housing Policy Matters for the Poor: Housing
Conditions in Latin America and the Caribbean, 1995–2006. Working paper. Washington, DC:
Inter-American Development Bank.
Sak, Hasim, Andrew Senior, and Françoise Beaufays. 2014. Long Short-Term Memory Based Recur-
rent Neural Network Architectures for Large Vocabulary Speech Recognition. Mountain View, CA:
Google.
United Arab Emirates (UAE) Drones for Good. 2017. “The UAE Drones for Good Award.” https://
www.dronesforgood.ae/.
United Nations. 2017. “Resolution Adopted by the General Assembly on 23 December 2016,”
71/25. New Urban Agenda. http://habitat3.org/wp-content/uploads/New-Urban-Agenda-GA-
Adopted-68th-Plenary-N1646655-E.pdf.
United Nations Programme for Human Settlements (UN-Habitat). 2011. Affordable Land and
Housing in Latin America and the Caribbean. Nairobi, Kenya: United Nations Programme for Human
Settlements.
World Bank. 2016. “Drones Offer Innovative Solution for Local Mapping.” http://www.worldbank.
org/en/news/feature/2016/01/07/drones-offer-innovative-solution-for-local-mapping.
Additional Reading
Aguilar, Adrián, Peter Ward, and C.B. Smith. 2003. “Globalization, Regional Development, and
Mega-City Expansion in Latin America: Analyzing Mexico City’s Peri-Urban Hinterland,” Cities 20
(1): 3–21.
Angel, Shlomo. 2000. Housing Policy Matters: A Global Analysis. New York: Oxford University Press.
Benítez, Griselda, Arturo Pérez-Vázquez, Martha Nava-Tablada, Miguel Equihua, and José Álvarez-
Palacios. 2012. “Urban Expansion and the Environmental Effects of Informal Settlements on the
Outskirts of Xalapa City, Veracruz, Mexico,” Environment and Urbanization 24 (1): 149–166.
Da Gama Torres, Haroldo. 2008. Social and Environmental Aspects of Peri-Urban Growth in Latin
American Megacities. New York: United Nations Expert Group Meeting on Population Distribution,
Urbanization, Internal Migration and Development.
Fay, Marianne, ed. 2005. The Urban Poor in Latin America. Washington, DC: The International Bank
for Reconstruction and Development; The World Bank.
Girshick, Ross, Jeff Donaue, Trevor Darrell, and Jitendra Malik. 2013. Rich Feature Hierarchies for
Accurate Object Detection and Semantic Segmentation. Berkeley, CA: University of California.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Missing ‘Middle Scenarios’: Uncovering Nuanced Conditions in Latin America’s Housing Crisis
43
Cityscape
Harvey, David. 1973. Social Justice and the City. Athens, GA: University of Georgia Press.
Karpathy, Andrej. 2015. “The Unreasonable Effectiveness of Recurrent Neural Networks,” Andrej
Karpathy blog, May 21. http://karpathy.github.io/2015/05/21/rnn-effectiveness/.
Mnih, Volodymyr, Nicolas Heess, Alex Graves, and Koray Kavukcuoglu. 2014. Recurrent Models of
Visual Attention. London: DeepMind Technologies Limited.
Olah, Christopher. 2015. “Understanding LSTM Networks,” Colah’s blog, August 27. http://colah.
github.io/posts/2015-08-Understanding-LSTMs/.
Ruprah, I.J. 2009. The Housing Gap in Latin America: 1995–2015. Washington, DC: Inter-American
Development Bank.
Secretaría de Desarrollo Social (SEDESOL). 2011. “Estado Actual de la Vivienda en México” Secretaría
de Desarrollo Social, Gobierno de los Estados Unidos de México. México City, México: Secretaría de
Desarrollo Social.
Shivdikar, Kaustubh. 2015. “Automatic Image Annotation Using a Hybrid Engine.” Paper pre-
sented at 2015 Annual IEEE India Conference (INDICON), New Delhi, India: December 17–20.
United Nations. 2010a. World Economic and Social Survey 2010: Retooling Global Development. New
York: United Nations.
———. 2010b. Home at Last? The State of the Homeless in Today’s Cities. New York: United Nations,
Department of Public Information and Non-Governmental Organizations.
Winchester, Lucy, and Raquel Szalachman. 2009. The Urban Poor’s Vulnerability to the Impacts of
Climate Change in Latin America and the Caribbean: A Policy Agenda. New York: United Nations,
Economic Commission for Latin American and the Caribbean.
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
44
Habitat III
This content downloaded from
155.33.31.132 on Sat, 09 Jul 2022 04:39:59 UTC
All use subject to https://about.jstor.org/terms
Article
Full-text available
Wide access to large volumes of urban big data and artificial intelligence (AI)-based tools allow performing new analyses that were previously impossible due to the lack of data or their high aggregation. This paper aims to assess the possibilities of the use of urban big data analytics based on AI-related tools to support the design and planning of cities. To this end, the author introduces a conceptual framework to assess the influence of the emergence of these tools on the design and planning of the cities in the context of urban change. In this paper, the implications of the application of artificial-intelligence-based tools and geo-localised big data, both in solving specific research problems in the field of urban planning and design as well as on planning practice, are discussed. The paper is concluded with both cognitive conclusions and recommendations for planning practice. It is directed towards urban planners interested in the emerging urban big data analytics based on AI-related tools and towards urban theorists working on new methods of describing urban change.
Book
Full-text available
This expanded and revised fourth edition of The Geography of Transport Systems provides a comprehensive and accessible introduction to the field with a broad overview of its concepts, methods and areas of application. Aimed mainly at an undergraduate audience, it provides an overview of the spatial aspects of transportation and focuses on how the mobility of passengers and freight is linked with geography. The book is divided in ten chapters, each covering a specific conceptual dimension, including networks, modes, terminals, freight transportation, urban transportation and environmental impacts, and updated with the latest information available. The fourth edition offers new material on the issues of transport and the economy, city logistics, supply chains, security, energy, the environment, as well as a revised content structure. With over 160 updated photographs, figures and maps, The Geography of Transport Systems presents transportation systems at different scales ranging from global to local and focuses on different contexts such as North America, Europe and East Asia. This volume is an essential resource for undergraduates studying transport geography, as well as those interested in economic and urban geography, transport planning and engineering. A companion web site, which contains additional material, has been developed for the book and can be found here: http://people.hofstra.edu/geotrans/.
Article
Full-text available
This paper discusses the evolution of housing conditions in urban areas of Latin America and the Caribbean from 1995 to 2006 based on data from household surveys done in 18 countries that comprise 95 percent of the urban population of the region. The results indicate that, on average, the proportion of urban households facing housing shortages is declining. This decline holds for households of all income levels, particularly those in the lower quintiles of the income distribution structure. The estimates made in this study indicate that in 2006 lack of infrastructure affected almost 19 million households. Further, about seven million households needed a new shelter and nine million needed significant improvements to their houses due to poor construction materials or overcrowding. Cross-country analysis shows that each country was facing a different combination of problems and was improving its housing conditions at a different pace, which indicates that it is highly unlikely that a ¿one-size-fits-all¿ solution exists.
Article
This paper proposes a critical analysis on the growing tradition of implementing public policies for the regulation of informal settlements in urban areas, aiming at contributing to a more adequate treatment of informal urban development in Latin America. The paper starts with a general discussion on the main causes of the phenomenon, stressing the need for preventive public policies to provide formal access to urban land and housing. Based on this analysis and on the main lessons from the Latin American experience we discuss the nature of existing regulation programs. As a conclusion, the paper argues that it is necessary to re-think the dominant legalization policies so as to promote, in an integrated manner, both security or tenure and socio-spatial integration.
Applying convolutional neural networks to large images is computationally expensive because the amount of computation scales linearly with the number of image pixels. We present a novel recurrent neural network model that is capable of extracting information from an image or video by adaptively selecting a sequence of regions or locations and only processing the selected regions at high resolution. Like convolutional neural networks, the proposed model has a degree of translation invariance built-in, but the amount of computation it performs can be controlled independently of the input image size. While the model is non-differentiable, it can be trained using reinforcement learning methods to learn task-specific policies. We evaluate our model on several image classification tasks, where it significantly outperforms a convolutional neural network baseline on cluttered images, and on a dynamic visual control problem, where it learns to track a simple object without an explicit training signal for doing so.
Article
Long Short-Term Memory (LSTM) is a recurrent neural network (RNN) architecture that has been designed to address the vanishing and exploding gradient problems of conventional RNNs. Unlike feedforward neural networks, RNNs have cyclic connections making them powerful for modeling sequences. They have been successfully used for sequence labeling and sequence prediction tasks, such as handwriting recognition, language modeling, phonetic labeling of acoustic frames. However, in contrast to the deep neural networks, the use of RNNs in speech recognition has been limited to phone recognition in small scale tasks. In this paper, we present novel LSTM based RNN architectures which make more effective use of model parameters to train acoustic models for large vocabulary speech recognition. We train and compare LSTM, RNN and DNN models at various numbers of parameters and configurations. We show that LSTM models converge quickly and give state of the art speech recognition performance for relatively small sized models.
Article
Can a large convolutional neural network trained for whole-image classification on ImageNet be coaxed into detecting objects in PASCAL? We show that the answer is yes, and that the resulting system is simple, scalable, and boosts mean average precision, relative to the venerable deformable part model, by more than 40% (achieving a final mAP of 48% on VOC 2007). Our framework combines powerful computer vision techniques for generating bottom-up region proposals with recent advances in learning high-capacity convolutional neural networks. We call the resulting system R-CNN: Regions with CNN features. The same framework is also competitive with state-of-the-art semantic segmentation methods, demonstrating its flexibility. Beyond these results, we execute a battery of experiments that provide insight into what the network learns to represent, revealing a rich hierarchy of discriminative and often semantically meaningful features.
Book
Latin American and Caribbean countries are the most urban in the developing world and have very high home ownership rates. However, many of the region's inhabitants are still poorly housed. This book examines three key contributing issues: high housing prices relative to family income, lack of access to mortgage credit, and high land prices.