+ All Categories
Home > Documents > Bright Lights, Big Cities? Review of research and findings...

Bright Lights, Big Cities? Review of research and findings...

Date post: 31-Aug-2020
Category:
Upload: others
View: 2 times
Download: 0 times
Share this document with a friend
46
1 Bright Lights, Big Cities? Review of research and findings on global urban expansion 1 David Mason, World Bank February 2017 1 This paper has been prepared as a background input to the World Resource Institute’s upcoming World Resource Report “Towards a More Equitable City.” The author is grateful for comments and guidance received from Mark Roberts, Chandan Deuskar, Anjali Mahendra and Sumila Gulyani.
Transcript
Page 1: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

1

Bright Lights, Big Cities?

Review of research and findings on global urban expansion1

David Mason, World Bank

February 2017

1 This paper has been prepared as a background input to the World Resource Institute’s upcoming World Resource

Report “Towards a More Equitable City.” The author is grateful for comments and guidance received from Mark

Roberts, Chandan Deuskar, Anjali Mahendra and Sumila Gulyani.

Page 2: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

2

Abstract

A common refrain among urban scholars and policy makers is that a majority of the world’s population is

lives in urban areas. However, the global extent of what is or is not “urban” remains unclear, often because

this definition varies by country. This makes it difficult to accurately compare urban expansion and

population growth trends between countries and globally. Recent empirical work has focused on

population-based and map-based techniques of estimating urban areas and urban expansion. This paper

provides an overview of these techniques as well as the main advantages and drawbacks in application as

well as gaps in knowledge and areas for future research and technical refinement. It finds substantial

divergences from previous estimates of global urban population as well as evidence that in many cases,

existing urban areas are expanding and becoming less dense. It also provides a brief review of recent

research utilizing remote sensing data to document the patterns and forms of urban expansion across the

world by region.

Page 3: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

3

AI Agglomeration Index

CIESIN

Center for International Earth Science Information

Network

DMSP-OLS

Defense Meteorological Satellite Program -

Operational Linescan System

GHSL Global Human Settlements Layer

GHSpop Global Human Settlements Population Grid

GRUMP Global Rural-Urban Mapping Project

GPW Gridded Population of the World

IMPSA Impervious Surface Area Map

MODIS Moderate Resolution Imaging Spectroradiometer

NASA National Aeronautic and Space Administration

SPOT Satellite Pour l'Observation de la Terre

TanDEM-X

TerraSAR-X add-on for Digital Elevation

Measurement

VIIRS Visible Infrared Imaging Radiometer Suite

WUP World Urbanization Prospects

Page 4: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

4

1) Introduction

Cities are home to a greater share of the world’s population every year, and according to the United

Nations, most people now live in urban areas (UN 2012). According to some, we have embarked on a new

“urban century” (Kourtit et al. 2014) where in the next two decades, the most populous continents, Asia

and Africa, are expected to become places with a majority of urban dwellers (Montgomery 2008). In

contrast to earlier views of urbanization as a disorderly, chaotic or maladaptive process, more recent

research has begun to improve our understanding of the processes and patterns behind how urban population

growth and expansion unfolds (Batty 2008). It is now clear that cities can hold certain kinds of persistent

social and economic advantages that rural areas do not have (Bloom et al 2008). Available case study

research shows that while most cities across the world have added population and built up area, they have

done so through different processes and different scales both between and within countries (Seto et al 2010).

There are also various institutional, environmental, economic and topographical features that shape the type

and form of urban growth (or decline). The confluence of a broad set of remote sensing data and improved

computing capacity have enabled researchers to develop multiple data sets, each utilizing different data

types, appropriate for different scales (city, regional, global), and with different purposes and intended

users.

Urbanization and economic growth appear to be strongly correlated and tend to occur

simultaneously (Henderson 2010; Glaeser and Maré 2011).2 Urban forms, particularly those that enable the

concentration and mix of people and firms, provide the basis for agglomeration economies (World Bank

2009; Glaeser 2011). In agglomerations, a diverse set of firms and labor are able to more efficiently sort

and match according to market needs (Venables 2010; Scott 2001). They also enable the rapid facilitation

of knowledge, ideas and learning, which further enhances innovation and growth (Jacobs 1969; Porter 2001;

Storper and Venables 2008). Agglomerations also enable economies of scale which make it easier and less

2 Africa appears to be a notable exception to this trend, where urbanization has occurred alongside lower levels of

per-capita GDP growth (cf Glaeser 2014).

Page 5: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

5

expensive for governments to provide networked goods such as roads and trunk infrastructure (Dobbs et al

2013; Libertun de Duren and Compeán 2015). Proximity is also particularly important for securing

economic security and social mobility. New migrants or informal workers often rely on social networks to

secure employment and access to housing and services (Mortensen and Vishwanth 1994; Massey 1990;

Reingold 1999; Helsely and Zenou 2014). Such networks are key gateways for economic mobility for

migrants and enclaves of ethnic or racial minorities (Portes and Landolt 2001; Edin et al 2001).

Despite the advantages that cities and urbanization can provide to social and economic

development, there are also costs. As incomes rise, urban dwellers tend to consume more and have a larger

natural resource footprint than people in rural areas, especially in developed economies (Hammer et al.

2007; Wackernagel 2006).3 Urban population growth also generates congestion costs or agglomeration

diseconomies, such as a greater incidence of air and water pollution, consumption of agricultural land,

traffic congestion, crime and the communicability of diseases (Ellis and Roberts 2015). In large cities, the

opportunities for employment may be constrained by the cost and time required to locate and commute to

such jobs.

New migrants and the poor often find difficulty in accessing public services and well-located

affordable housing given the comparatively higher costs of urban living (Molina et al 2002). A growing

body of work4 also examines how urbanizing areas may deepen or sustain social exclusion along different

lines including gender (Hagan 1998; Chant 2013), migrant status (Li 2006) and caste (Vithayathil and Singh

2012), among others. These present costs that can inhibit or attenuate the benefits that cities can provide

because they weaken the easy circulation of people and goods, limiting economic productivity, lowering

educational and health outcomes and reducing incentives for investment and innovation (World Bank

3 As McNeil (2000) wryly observes “Fast growing cities, like teenagers, have higher metabolisms than those that

have stopped growing” pg. 285. 4 See McGrahanan et al. (2016) for a timely review of these issues in relation to the Sustainable Development Goals

and World Bank (2013) for a broader treatment of the topic.

Page 6: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

6

2009). While all growing cities face these and other challenges, planning and strategic management of

urban population growth and expansion can reduce these negative impacts (World Bank 2013b).

While the many implications of future urban population growth are increasingly clear, the

definition and measurement of “urban areas” across the world is not (Cohen 2004; Satterthwaite 2010;

Roberts, et al. 2016; Deuskar and Stewart 2016). This is important for several reasons. First, without a

consistent definition of what is “urban,” it is difficult to accurately compare changes in both urban built up

areas and urban populations across time and space. The expansion of built up areas has direct impacts

quality of life, including increased commute times, diminished access to urban services, proximity to

agricultural land and green spaces. Second, the type and form of urban expansion matters because such

information can inform decisions by both policymakers and planners about how to ensure that space is used

both efficiently for land markets and the circulation of people and goods, but also improving social inclusion

and economic mobility. Third, these data can also be used in other areas to improve the assessment of

exposure to cities in terms of disaster exposure (McGranahan et al 2007), improving public health and

epidemiological applications for reducing disease incidence (Hay et al. 2004, Linard et al 2010), and

assessing hydrological and soil condition changes in food systems (Atzberger 2013).

As detailed in the UN’s World Urbanization Prospects (2014)5, definitions of urban areas differ

markedly across countries, covering a range of specifications and justifications. Out of 232 countries, 99

do not have an official definition of urban and 103 countries use a minimum settlement population standard

(Deuskar and Stewart 2016).6 In other countries, population density thresholds, settlements with economic

specializations, or the presence of services and infrastructure are used as definitions.7 As each country

5 Available at: https://esa.un.org/unpd/wup/ 6 For example, in Argentina this includes areas of 2,000 people or more whereas in Senegal settlements of 10,000

people or more are officially classified as urban. In Mali, urban areas were those with at least 5,000 people up

through 1987, then at settlements of least 30,000 people from 1998 onward. 7 The Philippines for example uses several requirements: density (at least 1,000 people/km2), designated

“administrative centers,” designated barrios with at least 2,000 inhabitants, barrios with at least 1,000 inhabitants

contiguous to an administrative center, and all designated municipalities with a population density of at least 500

people/km2. In Malaysia, urban areas require a population of at least 10,000, at least 60 percent of those over 10

years old engaged in non-agricultural activities and, finally, housing units with modern toilets (UN 2014).

Page 7: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

7

measures urban population differently, cumulative totals of urban population across regions - or the world

- lack reliability due to this inconsistency (Cohen 2004). Nonetheless, these figures are often cited as

evidence that more than half of the world’s population lives in cities, (UN 2008; Sudjic and Burdett 2007)

and as the basis for the UN’s estimation of future urban population growth in the coming decades (UN

2014). Without a consistent and comparable measure of urban areas and the populations they contain, it is

difficult to perform cross-country analysis of the relationship between urban population growth and

expansion and other areas of policy concern, such as land use planning, economic growth, climate change,

and disaster risk management, among others.

In recent years, advances in the processing of remote sensing technologies combined with

improvements in the quality and availability of population data has greatly improved the accuracy of

estimating the location and population of urban areas across the world (Potere and Schneider 2007; Seto et

al 2010). Over the last fifteen years, the utilization and refinement of these tools is improving our

understanding of the spatial and temporal dimensions of urban change, in terms of land use coverage,

population densities, urban expansion and the distribution of economic activity. Yet, there are limits and

tradeoffs to the current approaches and there is no universally accepted dataset or platform. Despite the

progress, there remains a lack of consensus on some basic questions, due, in part, to definitions and also

limitations and tradeoffs within the data sets constructed. Users have different needs; some may concentrate

more on land use and built environment measures, while others are more concerned with the distribution of

population, which leads to different definitions of urban areas. Finally, while the availability of data has

expanded rapidly, changes and improvements to the data sets are very frequent, which makes older sources

outdated and hinders comparison of observations across time.

While this work remains in a relatively nascent stage it is important because it will allow scholars

and policy makers a more consistent approach to understand the characteristics and qualities of urbanization

and land use changes, especially in places where it has previously been overlooked or poorly measured with

conventional approaches, such as is the case in South Asia and Latin America (World Bank 2015; Roberts

Page 8: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

8

et al. 2016). Additionally, the Sustainable Development Goal 11.3 aims to “by 2030 enhance inclusive and

sustainable urbanization and capacities for participatory, integrated and sustainable human settlement

planning and management in all countries” (UN 2016). A proposed measure for this goal is tracking the

land consumption rate to new population growth. For this purpose, a globally consistent measure and

methodology is necessary to effectively monitor the progress and outcomes of this indicator.

This paper aims to provide an introduction to research on urban expansion, in particular the datasets

and methodologies utilized and current trends observed at a global level, with examples from different

regions. This is important because while there is general agreement that the world is becoming more

“urban,” and that cities are expanding to accommodate this growth, the ways in which such change is

observed and measured depend on the map and population data used. A review datasets and approaches is

useful for informing policymakers and practitioners managing urban growth and expansion about the tools

available, their limitations and how they may be applied in policy settings at city, regional and national

levels. This paper will review different datasets and approaches to documenting and measuring urban

expansion, and offer caution on how these data may be applied for different purposes and suggest areas for

future inquiry.

Urban expansion refers to the growth in occupied built up area as part of urban settlements. “Urban”

has two related definitions. Under the first, approaches using satellite imagery or other remote sensing data

aim to classify urban areas based on a defined set of criteria for observed artificial built-up areas in relation

to other types of land coverage (forests, deserts, agricultural areas and so forth). The second draws from a

demographic approach; using built up area data as an input for estimating the distribution of populations in

and around known settlements in order to define urban populations by using criteria such as minimum

population, minimum population density or catchment area of travel to defined urban point. In each case

however, general, universal definitions of “urban” remain elusive. This is because what is “urban” exists

along a continuum; there may be a basic agreement that the central cores of very large cities have similar

Page 9: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

9

characteristics (a high concentration of buildings and residents), but there is a large variation in land use

types and intensities and patterns in population concentrations moving outward toward more rural areas.

This paper is organized into three sections. The first will review the remote sensing tools and population

sources that form the basis for current standardized measures of urbanization, distinguishing between

approaches that define “urban” through observable built up areas and those using population and density

characteristics of urban settlements. It will assess the assumptions, benefits and drawbacks to these

approaches. The second will survey findings from recent empirical work utilizing these tools to track

changes in urban form and populations from different regions, documenting trends and patterns observed.

A final section will summarize the conclusions and identify future avenues of research. Detailing the

explanations and causes for differences in growth patterns is critical for scholars and policymakers, though

it is beyond the scope of this paper.

2) Estimating Urban Expansion: Definitions and Methodologies

In recent years, the observation that most of the world’s population lives in cities has become an

increasingly common refrain among practitioners and urban scholars (UN 2014). However, this assertion

overlooks disagreements and differences of a precise definition of “urban” along with the methods for

identifying and comparing urban areas and urban populations across the world. In recent years there have

been rapid improvements in the quality and availability of remote sensing data and advances in its

application to estimate urban land coverage and urban populations. Much of this work has attempted to

draw estimates of built-up areas and combine them with spatially disaggregated gridded population data

sets. This section reviews the main sources of satellite imagery and the population databases used for

developing global urban population layer products.

Using Satellite Imagery to Map Urban Areas

There are several satellite-based maps available for examining global urban expansion by

estimating the area and changes in built environment based on analysis of multi-spectrum satellite

Page 10: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

10

observations (cf. Schneider et al. 2009). “Built environment” refers to land surface covered with human

constructed materials, such as roads, buildings and other artificial surfaces. The resolution of available

satellite maps influences how built environment surfaces are measured. For visible spectrum photographs

this ranges from coarse resolution (>500m; MODIS), moderate (30m; Landsat), high (~ 5m; Rapid Eye and

very high resolutions (<1m; Quickbird). There are several ways to identify and classify built environments.

This includes visual review, classification and tracing, semi-automated detection using both algorithms and

manual review and fully automated detection relying on computer-assisted review and classification of the

coverage types for each pixel.

Built environments may be classified according to manual tracing or analysis of multispectral

images8, or with algorithms that estimate probable built up coverage using specified parameters based on a

spectral analysis of the color, patterns and reflectivity of certain areas compared to others. Other sources,

such as nighttime lights (which dates back to 1992), capture the extent of artificial lights at a very coarse

(~0.5 km) resolution by specifying a threshold cut off measure for brightness separating urban and rural

areas.9 Different automation systems also use different definitions for urban features. For example, IMPSA

(Global Impervious Surface Area) categorizes urban areas as impervious surfaces, MODIS500 defines them

as areas of 1km2 with at least 50 percent coverage of artificial surfaces, GRUMP uses a nighttime lights

along with a secondary estimate of urban built up area (Potere et al. 2009; SEDAC 2016) This is an

imperfect process as distinguishing artificial from natural ground cover is subject to both errors due to

specification and also issues with image quality and clarity (as well as obstructions such as clouds or smoke)

(Small 2005). These different approaches to land use classification account for the substantial variation in

the estimates of the total area of built up urban land (See Figure 1 for comparison).

8 This includes images captured within the visible portion of the electromagnetic spectrum, but also higher

frequency infrared and radio waves 9 For example, observed brightness is assigned a scale numeric value, with those areas exceeding a certain number

deemed to be “urban” while areas with lower scores are rural.

Page 11: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

11

Figure 1: Urban Built-up Areas of Washington, D.C. – Baltimore Metropolitan Region, Select

Maps,10

Landsat 30m

Modis 500m

GRUMP 1km

IMPSA, 1km

There is no single ideal mapping platform for assessing global urban expansion. Satellite imagery

varies by the date of first observation, the frequency of regular updates or passbys, the resolution and spectra

of light frequencies captured, and cost. Other satellites, such as the German TanDEM-X also capture high

resolution, non-optical frequencies based on radar, infrared and laser reflections. These data can be used to

10 Drawn from Schneider et al. 2010, pg. 1741. The frame size for each map is identical. Yellow areas are classified

as “urban,” while darker green areas are rural or natural environments. Dark blue/black areas represent water bodies.

Page 12: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

12

refine the definitions of urban built up areas because they allow for distinguishing variations in the height

and shape of ground cover, such as buildings towers and poles (Taubenbock et al. 2011a). For global urban

mapping, cost and data coverage availability often requires the use of coarse resolution (250-1000m) or

moderate resolution (20-30m) mapping data (Mertes et al 2015). For automatic and semi-automatic

classification systems, resolution influences how different land coverage types are classified because a

single pixel unit may contain ambiguous or mixed coverage. At both coarse (>300m or more) and higher

resolutions (5m or less) it becomes difficult to identify and distinguish the surface variation within a given

pixel, so it can be difficult to classify both differences between built-up and non-built up areas, or to classify

different types of built-up areas at a broader, comparative scale.11

Recent research has also focused on identifying techniques to better estimate the extent of built up

area in order to enhance the calibration of population grids for regional and global comparison (Mertes et

al 2015; Tatem et al. 2004). This has involved comparing the estimates of urban land coverage generated

by automated classification systems from different low resolution maps with samples of higher resolution

reference locations and testing for agreement in terms of urban land cover estimates (Potere et al. 2009)12

Indeed, even among mapping data systems, there is considerable variation in the total amount of built up

area estimated due in part to image resolutions, conditions for distinguishing built-up areas from natural

environments and differing years of image collection. In general, the greater the resolution of maps the

higher the estimated total built-up area. For example, as Schneider (2007) observes, one early map, Vector

Map, drawn from a set of digitized and harmonized maps and navigational charts at a low resolution,

provides an estimated total urban surface area of just under 0.3 million km2. By contrast, GRUMP, which

11 Consider the great variety of colors, reflectivity, patterns and materials of building roofs, appurtenances and street

surfaces within a single city. In order to use these as markers or inputs for identifying built up areas in other cities

within the region or across the world, this “language” of colors and patterns for urban classification would have to

be refined considerably. It would have to both include a much wider variety of different land surface coverage types,

but also remain internally consistent enough so that these classifications are not applied to non-urban areas in error. 12 In the Potere and Schneider paper, reference points were drawn from Google Earth as well as other studies where

urban boundaries and land coverage were clearly delineated.

Page 13: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

13

still utilizes a coarse resolution of about 1km and a different algorithm, estimates this figure to be 3.5 million

km2 or 11 times larger (Potere and Schneider 2007).

In a comparison study of four early global map products,13 Potere and Schneider (2007) find that

there is some degree of inter-map correlation by region, with North America having the highest level of

convergence (r=0.90) and Asia the lowest (r= 0.63) with Europe, Latin America and Sub Saharan Africa

midway between these two. Such variation may likely be due not only to differences in the resolution of

the underlying satellite imagery, but also differences in automation algorithms and years of observation;

where rapid urban expansion may be recorded in different maps across two different time periods. They

also suggest that the strongest factor is likely due to differences in discerning urban land; “in the absence

of a clear set of definitions, each group constructs an implicit model of urban land that can be inferred from

their methodologies” (pg. 23). A follow up analysis (Potere et al. 2009) compared eight map products by

assessing their consistency in matching a sample of sites and defined cities as well as predicting the built

up area of these cities. As with previous studies by the same authors (Schneider et al 2009), the MODIS500

map was found to have the highest level of agreement with high resolution reference images, comparatively

superior detail and resolution and a lower level of omitted city errors. The authors use this map to estimate

Earth’s total urban surface area (around 2001) at 657,000km2 or about 0.44% of total land area, suggesting

that the total space occupied by urban land is rather small.14

Other work has focused on how to triangulate estimates of urbanized areas using multiple map

products (including those at different resolutions) which can be used as a baseline for monitoring future

urban expansion in a consistent manner. For example, Mertes et al. (2015) developed such a technique by

using a sample of East Asian cities which are otherwise subject to greater measurement error due to

13 The analysis includes Vector Map Level zero (VMAP0), Global Land Cover 2000 (GLC00), the History Database

of the Global Environment v3 (HYDE3), Global Impervious Surface Area (IMPSA), MODIS Urban Land Cover

and GRUMP 14 Angel et al. (2010) use MODIS500 to estimate current and future urban land coverage. They estimate this figure

to be 605,875 in 2010, with projected growth between 853,355km2 and 1.2 million km2 by 2020. This map classifies

“urban” as areas of 1km2 with at least 50 percent built up area.

Page 14: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

14

persistent cloud cover and a pattern of rapid urban expansion which has been difficult to consistently

identify as built up area.

An emerging strand of work utilizing multiple map sources has been completed as inputs for the

Global Human Settlement Layer (GHSL) (Pesaresi et al. 2013). The impetus for this work emerged

following the Indonesia tsunami of 2004, in an effort by aid agencies to obtain a large scale mapping

platform that could be regularly updated for disaster risk management planning and to monitor urban

expansion and refugee settlement issues (Pesaresi et al 2013). This layer is calibrated from seven high and

very high resolution satellite maps sources with lower resolution base layers as references for the period

circa 2010. The data also include four periods (1975, 1990, 2000 and 2014) which permit comparative

temporal analysis at the global level. This platform provides among the most comprehensive global level

estimates of urban coverage, with a recent update in 2015.

Other remote sensing techniques have utilized imagery of ambient night time lights from human

settlements. Nighttime lights data from the Defense Meteorological Satellite Program Optical Line Scanner

(DMSP-OLS) (archiving of images began in 1992) measure the extent of ambient light from human

settlements. The data allow for tracking both urban expansion by the location and extension of new lights

around cities, as well as changes of intensity and brightness of lights within cities, which suggest the

presence or absence of population concentrations (Zhang and Seto 2011). The resolution of these images is

coarser than other data sets (approx. 1 km at the equator), but they do allow for estimation of built-up areas

that can be cross-checked with other sources as is used by GRUMP (Balk 2009).

A key advantage of nighttime lights data is the broad scale and extent of coverage, which is useful

for mapping urban settlement expansion patterns at regional or global levels. Another advantage compared

to daytime, visible spectrum maps is that they also capture variations in brightness, especially within known

built up areas, which can help determine the relative intensity of economic activity and concentration of

population. There are some limits in the application of nighttime lights maps. For example, the resolution

is too coarse to reliably record very small settlements, and the data also suffers from a “blooming” or

Page 15: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

15

“overglow” effect where areas of intense brightness may obscure both the variation of brightness within a

settlement as well as its overall contour and bounds.15 Recent work has developed techniques to improve

the estimates of urban boundaries (Abrahams et al. 2016) and has begun to test newer, alternative nighttime

lights sets, such as NASA’s Visible Infrared Imaging Radiometer Suite (VIIRS). This instrument, originally

designed for recording cloud changes, surface temperature and fires (among other earth science

applications), can map nighttime lights at a higher resolution (742 m2 versus 5km2) which allows for

improved automated correction of blurring and glow effects (Shi et al 2014; Baugh et al. 2014).

Calibrated nighttime lights data also have other applications when combined with complementary

data. For example, there is evidence that the form and brightness of nighttime lights is correlated with GDP

and many economists have used this approach it to measure differences in the location and intensity

economic activity the urban level. (Henderson et al 2012; Chen and Nordhaus 2011; Mellander et al. 2013;

Ellis and Roberts 2015; Bundervoet et al. 2015).16 Figure 2 below shows sample nighttime lights images.

15 Corrections to this effect can diminish the intensity of lights from smaller settlements (Bloom et al. 2007). There

are several products with different applications; see Doll (2008) for a review. 16 Changes in the intensity and density of urban light patterns may also be proxy measures for population density

(Bagan and Yamagata 2015) and access to electricity (Doll and Pachuari 2010). Related applications include the use

of these data for estimating losses in the events of natural disasters (Gunasekera et al. 2015) estimating the extent of

global poverty (Elvidge et al 2009), assessing emissions of greenhouse gases (Elvidge, 1997), and the efficiency of

urban energy grids (Fragkias et al. 2016). There are a growing number of promising applications of nighttime lights

data as a complement to other geographic and socioeconomic data sources, especially as calibration methods and

corrections continue to improve.

Page 16: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

16

Figure 2: Select nighttime lights images of urban areas from South Asia17

Hyderabad, DMSP-OLS (2010)

Red indicates the brightest areas, green and blue

less bright. Black lines represent the city’s

administrative boundaries.

Central Indo-Gangetic Plain, VIIRS (2014)

The image shows a chain of cities (white areas) from Lahore at the upper

left hand corner stretching to the Delhi in the lower right hand corner

Linking Urban Areas with Population Data

As previously discussed, the UN World Urbanization Prospects data have the advantage of

simplicity because they draw on available census or survey population estimates. For each country,

estimates are based on available population counts and projections for urban and rural areas from observed

trends dating back to 1950. Across countries, the granularity of census data units or tracts varies widely and

some countries have not had regular decennial censuses, leaving large demographic data gaps.18 Changes

in urban populations are interpolated from available data, which may classify population boundaries for the

17 From Roberts et al. 2015, pages 62 and 65, respectively. 18 For example, Pakistan, one of the most populous countries on earth, completed its last census in 1998.

Page 17: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

17

same urban area in different ways in different years, making population trends inconsistent (Montgomery

and Balk 2011).19

The use of satellite imagery to distinguish land use types may allow for refinement of population

estimates according to built-up areas. Satellite images of built-up areas are overlaid with sets of gridded

population data based on recent or estimated population counts within administrative boundaries (see

Deichmann et al. 2001 for an overview of early approaches). These databases then estimate the likely

population for a given grid cell (e.g. 1 ha or 1km2) based on the built-up area and the known or estimated

population of the corresponding administrative area.

An early database, the Gridded Population of the World (GPW) developed by the National Center

for Geographic Information and Analysis at the University of California, Santa Barbara, with assistance

from the Center for International Earth Science Information Network (CIESIN) at Columbia University.

The most recent edition (version 4) covers 2000-2020 in five year increments20) uses population data from

the smallest administrative units available. Projections are estimated using census administrative units and

projecting a uniform population distribution across grids of 1km2, a technique called areal weighting

(Mennis 2003). Figure 3, drawing from Deichmann et al 2001, shows this using the example of Haiti. The

drawback to this approach is that populations are often not distributed evenly within cells which leads to

biases in country level urban population estimates where there are fewer census input units. However, the

advantage is that new, more granular census data can easily be integrated and population estimates quickly

refined (Doxsey-Whitfeld et al. 2015). For example, the first version, released in 1995, used 19,000

administrative units, where the third version, released in 2005, contained nearly 400,000 and the most recent

version, updated in 2015, has 12.5 million (SEDAC 2015).

19 For example, depending on available data, the population of a city may be arbitrarily reported to the UN in terms

of city proper, agglomeration or metropolitan region. 20 The figures for each of the years are extrapolated from 2010 census administrative unit inputs.

Page 18: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

18

Figure 3: Gridded Population Maps, Haiti, 2001

This map shows administrative boundaries and with a

1km2 grid matrix for an area of northern Haiti. Cells with

darker red have higher population densities.

This map shows the gridded population data for the

entire country, again where darker red areas

represent higher population densities.

Other products have attempted to correct for this with more sophisticated estimates of population

distribution. These include LandScan and GRUMP which use a dasymetric estimation approach through

integrating additional information to construct population distribution estimate within a given grid.21 For

example, LandScan, developed by the Oak Ridge Laboratories in Tennessee, estimates vegetation cover,

topographic variation and transportation corridors to calibrate population distributions within each cell

(LandScan 2016).22 However, in contrast to GPW, it uses a proprietary method for assigning population

which is regularly updated, making reliable population comparisons across time periods difficult. Similarly,

GRUMP uses a set of 55,000 settlement points with populations of at least 1,000 people and draws on an

urban extent layer drawn from multiple sources including nighttime lights data from DMSP-OLS to

calibrate the population weighting (Bloom et al 2007).

21 This approach has also been developed and utilized for the most recent GPW set. 22 For example, in a given cell, a greater share of population may be distributed closer to roads and intersections,

rather than in forested or undeveloped areas.

Page 19: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

19

Worldpop and GHSPop provide gridded population data based on census inputs and permit global

level comparative analysis of urban populations. WorldPop uses available census and population data,

nighttime lights data, and several other layers of data to map urban extents, along with a machine learning

algorithm to model population to 100m2 cells (Stevens et al. 2015). The data currently exist for several

dozen countries with more being added regularly in Asia, Africa and Latin America.23 The Global Human

Settlements Population Layer (GHSPop) was developed by the European Commission’s Joint Research

Center and (JRC). Recent versions draw population data from CIESIN’s GPW and now utilize the GHSL

layer for the extent of urban coverage to distribute population in urban areas four periods beginning in 1975

at a resolution of 250m (Freire et al. 2015).24 The global coverage, depth of data, ease of access and

refinement have increased the attention and interest in their utilization for estimating urban expansion

(Deuskar and Stewart 2016).

Demographic Approaches to Defining Urban Areas

The previous section examined how urban areas are measured in terms of the surface coverage of

built up areas and the distribution of populations within them. Another approach defines urban areas in

terms of the population characteristics of settlements, such as minimum total population, minimum

population density and or distance from a known central point within a settlement. One of these methods,

the Agglomeration Index (AI), (Chomitz, et al, 2005 Uchida and Nelson 2010, World Bank 2009) uses each

of these three conditions to construct a measure of urban agglomeration. Agglomerations represent

concentrations of workers and firms that derive mutual benefit and increasing returns to scale from close

proximity in the same geographic space (Marshall 1961; Storper and Venables 2004). Agglomerations

provide a functional concept of urban areas by defining them in terms of a concentration of population

23 Older data exist for additional countries, though the estimation technique used is different compared to the current

edition. 24 CIESIN is also developing a gridded population product with Facebook that utilizes the company’s image

recognition algorithm to process images with built up layers at very high resolution (50cm) for 21 countries.

http://blogs.ei.columbia.edu/2016/02/22/working-with-facebook-to-create-better-population-maps/

Page 20: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

20

around economic activities.25 Rural areas, by contrast, are characterized by more dispersed populations and

points of exchange. The AI applies a population condition (settlements of at least 50,000 people) a density

threshold (at least 150 people/km2) and the catchment of a 60 minute travel time to city center points

identified on GRUMP or LandScan maps. Travel time provides an important qualifier because it is a proxy

measure for some level of labor market participation within the agglomeration. However, part of the

weakness for using the AI to track changes in urban populations over time is the lack of broad availability

and timeliness of travel time data.26

The “Cluster” method, introduced by the European Commission, represents a similar, population-

based approach but without travel catchment times. It was initially applied to European cities but has since

been updated and expanded (Dijkstra and Poelman 2014; Deuskar and Stewart, 2016; Roberts et al., 2016).

The Cluster method takes population layers from a gridded population distribution map (such as WorldPop

and GHSPop) and categorizes them as clusters depending on population size and density. Urban clusters

composed of contiguous gridded population cells with a minimum population of at least 5,000 and a density

of 300 people/km2. Within this set, ‘high density’ clusters are defined as clusters of built-up area with at

least 1,500 people/km2 and a settlement size of at least 50,000 (Dijkstra and Poelman 2014). All other cells

are classified as “rural.” A benefit to this method is that thresholds can be adjusted according to need for

different types of comparative analyses as it arranges the values of gridded population cells according to

specification. However, the selection and use of different population condition thresholds requires an

appropriate analytical justification.

Revisiting Urban Population Estimates Work

Recent analytical work has compared the estimates of global urban population using the AI and

Cluster methods by drawing from WorldPop, GHSPop, and LandScan gridded population datasets with

25 Or as Bertaud (2003) more succinctly observes “the raison d’etre of large cities is the increasing return to scale

inherent in large labor markets” (pg. 1). 26 The layer recording city center points, from which the travel time catchment is calculated, had been based on the

original points from 2000 and was updated in Roberts et al. (2016).

Page 21: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

21

figures proposed in the WUP. Roberts et al. (2016) compare these datasets for a global sample of countries

using the most recent data (circa 2014) utilizing both the Cluster and AI methods to derive estimates of

national urban population shares by region. They find that each approach estimates a higher proportion of

urban population in most regions than does the WUP, most notably in South Asia, East Asia and Sub

Saharan Africa. GHSPop estimates for South and East Asia are even higher; suggesting that about 80

percent of the population of these regions is urban (versus approximately 57 percent in WUP and 64 percent

in WorldPop, respectively).27 Figure 4, drawing from Deuskar and Stewart (2016) demonstrates the

difference in urban area estimates between WorldPop and GHSPop when the same density and population

thresholds under Cluster method are applied to Jakarta, Indonesia. It shows that Worldpop estimates a much

larger urban area overall than does GHSPop, while GHSPop identifies a greater extent of high density

clusters.

Figure 4: Comparison between WorldPop and GHSPop using the Cluster Method, Jakarta28

WorldPop, 2015

27 Among the two population distribution inputs (WorldPop and GHS Pop), this difference is due to how population

is assigned within each cell; Worldpop uses a more conservative specification which tends toward a more diffuse

population distribution with cells. It also draws from lower level census data with more granularity and specificity,

whereas GHSpop assigns the total population to larger (1km2) cells and often draws from less granular population

data. 28 From Deuskar and Stewart, 2016

Page 22: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

22

GHSPop, 2015

The Cluster Method classifies cells within a gridded population distribution map according to density and then clusters

them by areas of similar density. High density clusters (shown in red) are contiguous cells of at least 1,500 people per

km2 with a total population of at least 50,000. Urban clusters (shown in yellow) are defined by contiguous cells with

a density of at least 300 people per km2 and a population of at least 5,000.

WUP data has also suggested that Latin America is much more urban than would otherwise be

assumed based on GDP-per capita and agricultural contributions to GDP. Roberts et al. (2016) applied the

AI and Cluster methods utilizing these three gridded population datasets for the region and found the share

of urban population is estimated to be much lower and more in line with global averages and the expected

relationship. The authors find that the AI method generates similar estimates with either the WorldPop or

GHSPop datasets and is slightly more robust than the cluster method (likely due to the inclusion of the

central city travel metric). This suggests that the choice of gridded population data (LandScan, WorldPop

and GHSPop) is a key determinant in estimated urban population outcomes and additional comparative

work utilizing these data is required to better understand these differences.

This work highlights both the advantages and limits of current approaches to measuring global

urbanization. The availability and variety of mapping data from remote sensing sources has allowed

scholars and planners to better identify the extent of built up areas. However, there remain several limits to

Page 23: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

23

tracking urbanization in terms of built up area due to differences in availability and granularity of data and

also differences in how to reliably classify different types of built up areas observed, particularly along the

urban periphery (rather than central city areas) where expansion is occurring. Compared to standard census

or gazetteer sources, gridded population layers have improved estimates of how populations are likely to

be distributed in and around known urban settlements and rural areas. The cluster and AI methods have

also provided a consistent and globally comparable way of classifying of urban areas based on a functional

definition of urban population concentrations (though the population conditions they use are somewhat

arbitrary). However, there lacks an explicit spatial dimension to this approach. We do not know for example,

how clusters and agglomerations overlap with built-up areas, administrative boundaries or artificial land

cover changes at global level.

Remote Sensing to Track Land Uses

As previously mentioned, medium and coarse resolution mapping data such as MODIS and

nighttime lights have been used to distinguish “urban” and “rural” areas based on identifying and

distinguishing artificial built up area from agricultural or natural land coverage. These maps, however, do

not show differences particular types of built up areas within urban spaces. For example, central city areas

with high rise buildings are clearly distinct from low rise residential suburbs, expansive industrial parks or

informal settlements, each of which may be located within the same functional urban area (or adjacent to

each other) but do not share the same distribution of population, economic activity, let alone land cover

pattern. These distinctions are important for better understanding city-level issues related to access to

services and employment, economic activity, transport and mobility and urban land markets and

administration, among other issues.

The quality and increasing availability of high and very high resolution (< 2m) satellite imagery

from multiple sources has also provided data for case specific analyses of urban fabric and land uses

(Graesser et al. 2012). These approaches utilize algorithms that classify map data either by shapes or by

vectors in order to distinguish colors, textures and patterns that are associated with certain types of land

Page 24: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

24

uses or neighborhoods, such as informal settlements (Hoffman et al. 2008). This method is further

strengthened by the inclusion of “ground truthing” photos taken by observers to improve the reliability of

the algorithm by triangulating the satellite maps with on-the-ground observations of building types, colors

and materials (Kim and Liu 2004, Lozano-Gracia et al. 2016). As additional data are included and the

algorithm conducts recursive tests, predictive accuracy is improved through a process of machine learning.

This process improves the identification of specific features, such as roads, vegetation but also building

types, heights and even distinguishing between informal and informal settlements (Kuffer et al 2016

Henderson et al 2016). There is ready value for this approach in mapping the location and extent of different

land use types within the larger urban fabric that would otherwise be missed by medium or coarser

resolution maps, such as hidden or “pocket” slums and informal settlements built with non-standard

materials (Kim and Liu 2004) or the location and accessibility of parks, green spaces and urban services.

3) Urban expansion: Trends and patterns across the world.

While researchers are improving techniques for consistently measuring the global urban footprint

and its evolution of time, a number of studies have examined urban expansion in cities and regions

throughout the world using satellite imagery. Due to the limited amount of comparable data across time

periods, much of this work utilizes purposive sample of city-level case studies and attempts to discern

patterns or typologies of urban change (Angel 2005; 2010). Since 2009, the World Bank has developed

several regional studies utilizing medium- (MODIS) and coarse (DMSP-OLS nighttime lights) resolution

satellite imagery to map urban expansion changes, especially in Asia (see, in particular, World Bank, 2015,

and Ellis and Roberts, 2015). This section provides an overview of findings from these and other recent

studies.

Despite disagreement on the precise amount, there is a consensus that urban areas, or at least built-

up areas, across the world are expanding (Seto et al. 2011; Angel 2005). In a recent meta-analysis, Seto et

al. (2011) reviewed 326 studies using remote sensing imagery and found all regions over the period of

Page 25: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

25

1970-2000, urban land expansion rates are equal to or greater than population growth, with the greatest

gains in India, China and Africa. In these areas, they estimated a global increase 58,000km2 of urban for

the period 1970-2000, with India, China and Africa having the fastest rates of urban expansion. In coastal

China, for example, annual urban expansion rates exceeded 13 percent over this period, while in areas

where expansion has not been as rapid, such as North America, growth rates were around 3 percent per

annum. Extrapolating from these patterns they use MODIS land cover maps, population and economic

growth projections to estimate that by 2030 the world will add approximately 1.5 million km2 or roughly

twice the amount of urban area estimated for 2001 (Schneider et al. 2009).29

Angel et al. (2005) find evidence for declining average population densities in cities across the

world. They utilized a global sample of 120 cities with populations of at least 100,000 and used Landsat

(30m) maps from 1990 and 2000.30 While they also find that cities in developing countries are on average

more dense than those in developed countries, in all sampled cities, population density declined on average

by about 2 percent per year or from an average of around 144 persons/hectare to about 112 persons/hectare

(pg. 57). They also find significant positive relationships between increasing density decline and cities with

rapid income growth, high initial densities and few physical or topographic constraints.

An emergent area of research attempts to classify differences in growth patterns within individual

cities to develop typologies of urban form and density. Schneider and Woodcock (2008) observe similar

patterns with a sample of 25 cities using a similar methodology drawing from census figures and Landsat

data from between 1990 and 2000. While all cities are marked by an overall decline in average density,

they identify four types of growth: about half of the cities are classified as “low growth” where more than

29 In each case, “urban” is defined in terms of artificial built environment categorized using medium/coarse

resolution satellite maps. 30 Locations were validated using Google Earth central reference points, but utilized available census data rather

than gridded population data. The study also draws is drawn from a previous sample of 3,945 cities with a

population of at least 100,000. A parallel analysis of 30 cities from across the world over the period 1800-2000

utilized historical maps and gazetteer sources also found a tendency for density to decline by about 1 percent per

year.

Page 26: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

26

half of converted land is infill development, suggesting a more compact growth pattern. 31 “High growth”

cities are characterized by a majority of land converted into scattered, non-contiguous development.32

American cities are classified as “expansive;” low density, with a large footprint and dispersed or

fragmented built up areas. By contrast, Chinese cities have followed a “frantic” growth pattern which

includes fragmented development, but at a consistently higher density.

East Asia: Urbanizing with density

The World Bank (2015) drew from MODIS500 maps and WorldPop data from 2000 and 2010 to

analyze urban expansion changes in East Asia.33 The report identified and assessed 869 urban areas which

consist of the built up area of cities with populations of at least 100,000.34 In contrast to the findings by

Angel et al. (2005), the report identified a pattern of both spatial expansion and densification in urban areas

of 100,000 people or more (Schneider et al 2015). The report finds that about 36 percent of the region’s

population lives in urban areas of this size. Over this time, urban expansion averaged 2.4 percent per year.

The highest rates of built up area expansion occurred in upper income countries, while urban population

growth averaged 3 percent but with higher rates in lower income countries. Mean urban population density

also increased 0.5 percent to 5,776 people per km2 in 2010.35

The study found different patterns across countries. Some two thirds of new urban land growth in

the entire region occurred in China alone (23,600 km2). Urban Indonesia became denser, with a density

increasing 27 percent to 9,400 people per km2 or adding only about 40m2 of urban area per person

(compared to 260m2 in China). The region’s eight “megacities” (built up urban areas with greater than 10

million people) had lower rates of population growth and land consumption than did small (<1 million) and

31 These include Guadalajara, Curitiba, Cairo, Nairobi and Ahmedabad. 32 Examples include Brasilia, Ankara and Bangalore. 33 This includes China, Indonesia, Vietnam, the Philippines, Japan, Korea, Korea DPR, Mongolia, Thailand,

Cambodia, Lao PDR, Myanmar, Malaysia, Timor-Leste, Papua New Guinea, Singapore, Taiwan and Brunei

Darussalam. 34 Urban areas are defined as the contiguous urban land in and around known settlements with a population

threshold of at least 100,000 people. Commuting or density conditions are not used to define this sample. 35 If China is excluded, mean urban population density rises to 6,600/km2

Page 27: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

27

medium (1-5 million) urban areas. Among these, megacities in lower middle income countries (such as

Manila and Jakarta) had the largest urban populations, but most new population growth occurred in smaller

urban areas. However, upper-middle income countries, with a more even distribution across different sized

urban areas saw the greatest growth in the largest cities (e.g. Shanghai).

These findings are in line with study of 142 Chinese cities over a longer time period (1978 to 2010)

which finds differences in the expansion and population growth of Chinese cities by size class (Schneider

and Mertes 2014).36 For example, the authors find that existing large (> 1 million population) cities,

especially in the coastal region were the main consumers of new land, most of the population gains over

this period occurred in neighboring smaller cities, often within the same agglomerations surrounding large

cities. This difference may be due in part to restrictions which deter rural migrants from large cities, but as

others have found, rural-urban migration has accounted for as much as 56 percent of urban population

growth in China from 2000-2010 (World Bank and DRC 2014).37

The World Bank report also highlighted a regional trend toward metropolitan fragmentation,

defined as urban areas where no administrative district contains more than 50 percent of the built up area.

Excluding in China, 41 percent of the region’s 869 identified urban areas are “fragmented,” (examples

include Metro Manila and Tokyo, Japan). The second most common expansion pattern consisted of

“spillover” urban areas - those where one administrative district has more than half of the built up area, but

less than 100 percent (examples include Hangzhou in China and Bandung in Indonesia). Large urban areas

– “megacities” (>10 million) _are entirely fragmented, with greater shares of spillover observed among

those between 0.5-5 million, while most of the smallest urban areas in the sample are still contained within

one administrative boundary. The expansion of urban built up areas across administrative boundaries has

36 However, where the World Bank study finds that urban land and urban population growth in China from 2000-

2010 are roughly equivalent (~3.2 percent annually), Schneider and Mertes (2014) find that urban land from the

sample of cities tripled, while urban populations only doubled, suggesting a trend toward less density in the cases

examined. 37 For example, the World Bank (2015) estimates that between 2000 and 2010 China added some 200 million urban

residents.

Page 28: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

28

important implications for intergovernmental coordination for service delivery as well as ensuring access

to housing and employment across diverse and often competing jurisdictions.

South Asia: Natural Growth and Urban Fragmentation

The World Bank’s South Asia Urban Review (Ellis and Roberts 2015) tracked changes in urban

expansion patterns using both nighttime lights data and the AI rather than WorldPop and optical satellite

imagery. In contrast to the patterns of urban expansion in East Asia being driven by rural-urban migration

(Schneider and Mertes 2015; Zhang and Song 2003), urban expansion in South Asia appears to be driven

more by natural population growth and the reclassification of administrative boundaries (Ellis and Roberts

2015). 38 In contrast to East Asia, particularly, China, the study finds that changes in the national level urban

population share in South Asia is often exceeded by the urban population growth rate.39 This means that

while urban areas are gaining population, the population growth in rural areas is not being offset by rural-

urban migration. For example, in Pakistan, in the period 1981-1998, just 26 percent of urban population

growth was from rural to urban migration (Karim and Nasar 2003).

This study uses both the AI to measure urban areas in terms of population conditions and also

applies nighttime lights data to track the extension of urban areas in the region. DMSP-OLS nighttime lights

data used by the report to define urban footprints show that between 1999-2010 urban areas in the region

as a whole, driven mostly by India (11 percent), grew at about 5 percent a year, more than double the urban

population growth rate over this period.40 In Pakistan and Sri Lanka, urban expansion rates were modest;

4.7 and 1 percent, respectively. Major cities in the region such as Delhi, Mumbai, Hyderabad and Colombo

(Sri Lanka) each experienced faster population growth in surrounding districts than in the city

administrative areas themselves. A consequence of this is a pronounced pattern of built up areas

38 This region includes Afghanistan, Pakistan, India, Bangladesh, Nepal, Sri Lanka, Bhutan and Maldives. 39 These estimates are derived from UN population data. 40 As discussed earlier, nighttime lights are used to estimate urban land coverage here by assigning a brightness

threshold and classifying urban and rural areas based on brightness measurements

Page 29: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

29

overflowing administrative boundaries. Using an analysis of Landsat imagery, the Indian Institute for

Human Settlements (2011) found that in Chennai and Kolkata for example, a greater share of the built up

footprint is outside of the principal administrative area than is within it. The result of this pattern is an

increase in multicity urban agglomerations (defined as continuously lit urban areas which contain at least

two cities with populations of at least 100,000), which have expanded at an annual rate of 8.6 percent.41

The result is a blossoming of new, large agglomerations around major cities (such as in Coimbatore, India)

and the linking of multiple cities in a continuous agglomeration of lighted urban area stretching hundreds

of miles from Delhi to Lahore and containing roughly 73 million people (Ellis and Roberts 2015).

Urban Density: City Level Findings

Urban density is an important indicator of population distribution within a city and its proximity to

services and jobs. However, simple density measures (population per unit of built up area) do not capture

variations in internal population distribution according to land use types, nor proximity to or concentrations

of economic activity.42,43 Certain types of urban density provide economic and mobility advantages by

reducing the cost of movement of people and goods. Density gradients have long been utilized for assessing

the concentration of people within cities (Clark 1951). In a classic monocentric city model, density is very

high in the central business district and declines at an exponential rate for each distance unit further outward

due to the increased transportation costs and the declining premium of land rents from these areas (Muth

1965; Alonso 1969).44

Available evidence suggests that this basic framework of a smooth monotonic decline in density

from a central point still describes the density form of most cities. Malpezzi and Bertaud (2003; 2014)

41 Multicity agglomerations here consist of a two cities with at least 100,000 people living it its administrative

boundaries and which share a continuous built up space. In 1999 there were 37 such agglomerations with an average

4 cities within the boundaries. In 2010 there were 45, with an average city count per agglomeration of just under 5. 42 For example, the density of the borough of Manhattan, in New York City is roughly twice that of the city as a

whole (ACS 2008). 43 For discussion of alternative measures of urban form, see the work of MIT’s City Lab

http://cityform.mit.edu/projects/metropolitan-form-analysis-toolbox-for-arcgis 44 Scholars have also used interpretation of satellite imagery to better define and classify polycentric “mega-regions”

in and around the world’s largest cities, see Taubenbock et al 2014 and Taubenbock and Wiesner 2015.

Page 30: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

30

utilizing a global sample of 40 and 57 cities respectively, estimate population density gradient, from across

the world, using data from 1990-2009. They find that most cities generally fit the negative density gradient

predicted, but there are a number of cities (such as Moscow, Johannesburg, Brasilia and Seoul) which do

not fit the expected gradient (density is either relatively stable or actually increases the further one travels

from the city center). They suggest this may be due to differences in regulatory environments, but current

measures of these factors to date are imperfect and not uniform.

In a sample of 84 cities, Richardson et al. (2000) find that density levels in differences in population

density of central city areas with surrounding urban areas, finding that average density levels in cities in

developing countries tend to be higher and density gradients less steep than in developed countries. This

finding is supported by Huang et al.’s (2007) assessment of the manually traced urban areas drawn from 77

global cities. They observe “the compactness, density and regularity of urban areas in developing regions

generally exceed levels throughout developed countries” (pg. 11).45 This finding is intuitive; in dense cities

where incomes are low we would expect the amount of space per person to be lower than in less dense,

higher income cities. However, in part due to data limitations, these studies combined only include two

cities (Abdijan and Dakar) from sub-Saharan Africa. This suggests the trend toward compactness is not

uniform and that other factors, such as changes income levels, may be linked to the forms of urban

expansion observed in different regions.

Latin America: Losing Compactness

The trend toward declining urban density also appears to be occurring in Latin American cities.

Using Landsat images of 10 major Latin American cities and then manually tracing and cross checking the

urban boundaries, Inostroza et al. (2013) assessed changes in urban form and density between 1985 and

2010.46 While existing city cores still constitute 92 percent of the urbanized area in the sample, little infill

45 This analysis utilized a technique of manually tracing of built-up boundaries and transferring them to GIS

shapefiles. 46 Census data at the lowest available administrative level were used as population inputs.

Page 31: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

31

development has accommodated population growth over this period. They find only about 30 percent of

new development took place within infill areas; only two Colombian cities, Bogotá and Cordoba grew with

infill development, the former increasing population density by 27 percent. Approximately 21 percent of

total expansion could be characterized as “leap frog” development, driven by Montevideo and Buenos

Aires. The remaining 49 percent of built up area growth occurred along existing primary radial roads (most

common in La Paz, Asunción, Brasilia and Santa Cruz). Fringe areas were found to be growing at a rate of

11 percent annually, whereas core areas expanded at just 2.9 percent.

A recent World Bank report on urbanization in Central America (Maria et al. 2015) uses Worldpop

to identify 167 agglomerations with populations greater than 15,000. The report identifies two trends in the

region; a greater urban primacy than official statistics suggest, and a pattern of low density expansion. For

example, the Managua agglomeration holds 55 percent of the Nicaraguan population; twice the figure of

27 percent reported by official sources. In Costa Rica, the San Jose agglomeration contains an estimated 85

percent of the country’s urban population. Between 2000 and 2010, most population growth concentrated

in secondary cities of populations between 15,000-100,000. As with other regions, the annual rate of urban

expansion has exceeded annual population growth.47 This is compounded by the finding that 43 percent of

the agglomerations now span three or more municipalities, suggesting highly fragmented urbanization.

As with other studies - with East Asia being a clear regional exception - recent data again show a

pattern moving toward lower urban densities.48 Large urban centers hold the majority of urban dwellers,

but there is a trend toward growth in secondary cities and along the urban periphery. As urban

agglomerations expand and consolidate smaller settlements, these areas also become more fragmented, as

is the case in Asia. Bogotá stands in contrast as an increasingly dense city, marching against the general

trend in the region toward greater urban space per-capita through low density or leapfrog expansion.

Africa: Dense cities, but Few and Far Between

47 The total built up area in the region since 1975 has tripled, growing at an average annual rate of 7.5 percent. 48 There are also important exceptions at the country level, such as patterns observed in Pakistan

Page 32: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

32

Urbanization in Africa has attracted growing attention in recent years as the region is undergoing

urbanization without the same per-capita economic gains historically observed in other regions (Fay and

Opal 2000). Linard et al. (2010) utilized Afripop data (part of the Worldpop platform) to assess the

distribution and accessibility of settlement patterns throughout the continent. In Northern and Southern

Africa, settlements tend to be spatially clustered more closely, suggesting closer proximity to major urban

centers, especially in higher GDP countries such as Libya, Egypt and South Africa.49 In these areas, 90

percent of the population live on 11.5 percent and 7.6 percent of the land versus the Central, East and West,

where the same share of the population occupies between 23 and 36 percent of land, providing a very

general contrast of settlement concentration versus dispersion.50 Settlements in the Central, East and West

regions are marked by greater dispersion across the land area where the average distances from large cities

tend to be greater, again illustrative of lower overall concentration of population in and around major cities

in these regions.

New research using remote sensing data is also beginning to describe the form of major African

cities. Large cities tend to share the same negative density gradients observed in other cities across the

world, but cities in the region standout for the comparatively scarce concentration of economic activities in

city centers (World Bank 2016). Nighttime lights data reveal that the concentration of economic activity in

African cities is low; at 5km from the city center it is around half the intensity observed in cities of other

regions with similar population densities. However, Henderson and Nigamaulina (2016) find that while

African cities have very high peak population densities compared to other regions, they also have much

steeper density gradients, with density declining 14 percent per kilometer from the city center (versus 9

percent in other regions). This suggests that city centers are dense, but there are few pockets of contiguous

density in urban cores.

49 However, this correlation is not statistically significant. 50 This difference is also reflected in average travel times to cities of at least 50,000 people. In the North and South,

this averages around 2 hours versus nearly 5 hours in the Central and Eastern parts of Africa.

Page 33: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

33

Urban expansion trends, however, do not show high levels of infill development. In Maputo and

Harare for example, 30 percent of land within 5km of the city center is vacant (World Bank 2016). In an

analysis of built up area changes in 21 cities from 2000-2010 based on Landsat (30m) and SPOT (2.5m and

10m) satellite imagery, the World Bank (2016) finds between 44 and 77 percent of new urban growth

occurred at or beyond the edge of existing urban areas.51 In Bamako and Maputo for example, half of new

growth occurred in a leapfrog pattern of patches of non-contiguous built up areas. Additionally, the rate of

leapfrog growth has increased in all but two cities (Windhoek and Addis Abbaba) between the periods

1990-2000 and 2000-2010. This demonstrates that cities in the region show a trend toward low density and

fragmented development.

Urban Decline and Population Dispersion

While there is general consensus that the majority of developing country cities have experienced

expansions in both area and population, this is not universally the case. An emergent literature on

“shrinking cities” has highlighted some of the trends and implications for this type of urban change and its

relationship to economic competitiveness (Pallagst et al. 2009). Over the last 50 years, 450 cities with

populations of over 100,000 have seen population decline of at least 10 percent (Oswalt and Reinitz 2006).

Examples of this trend occur in developed countries such as the United States, Canada and Australia as well

as in former command economies in Eastern Europe and the former Soviet Union, where, in each case,

overall urbanization levels are historically high.52

Ukraine provides a unique case study. Since the 1990s, the average annual population growth rate

in the country has been negative. Despite this, the country has had the fifth greatest increase globally in

built up 1975-2014 according to GHSL (Ionkova and Sulukhia 2015).53 Furthermore, night time lights

51 This is based on a dataset assembled by Baruah (2015) 52 China provides an interesting example where a large construction boom and investment in “new towns” has

provided a large tracts of unoccupied apartments and vacant commercial buildings. See Shepard (2015) for a

detailed case study. 53 The area totaled 21,222 km2, behind only China, the United States, India and Russia and ahead of Indonesia,

Brazil, Turkey and South Africa.

Page 34: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

34

analysis of urban expansion and brightness reveal declining urban footprints and nighttime lights brightness

in the formerly urban east, while western cities became larger and brighter (World Bank 2015). The report

also finds that economic growth is closely correlated with cities that have both expanded their footprints

and have central cores that have become brighter. By contrast, cities with declining economic

competitiveness have dimmer cores, and those with the lowest economic competitiveness actually have

experienced dimming of both cores and peripheries, but an overall increase in the total built up area (pg.

184). For scholars, the study raises additional questions about the relationship between urban expansion

and economic competitiveness as these patterns contrast with the findings in India using a similar analysis

of nighttime lights and urban GDP data (Tiwari et al 2016). For policymakers, these findings call attention

to questions about how to manage infrastructure provision in cities experiencing net outmigration or

negative population growth. The case also demonstrates how nighttime lights data can be an important

complement in establishing contrasts between the extents of built up areas and the concentration of

economic activity within them – which may not otherwise be apparent.

Summary

Large cities in poorer countries also appear to have greater population density than those in

wealthier countries, but nighttime lights data demonstrate that they are not necessarily witnessing increases

in the intensity of economic activity in central city areas. There are some exceptions; Bangalore for

example, maintained brightness levels in central city areas (Ellis and Roberts 2015) and in Central America

the Guatemala City agglomeration increased its share of the total brightness observed for the country

between 1996 and 2010 (Maria et al. 2016). In parts of Eastern Europe, population movements and

structural changes in the economy appear to be reflected in a general dimming of declining urban cores and

in the case of Ukraine, increased, but dim, growth of the periphery. The comparatively rapid expansion of

medium sized or secondary cities also underscores the need to understand how these cities can plan for

future growth, along with establishing what tradeoffs new migrants are making by locating to these places,

rather than larger primary cities. While this paper does not offer explanations for these patterns, attention

Page 35: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

35

should be given to differences in land administration regimes, structural economic shifts and urban and

metropolitan governance institutions and arrangements as key factors.

4) Conclusion and Future Directions

This paper has reviewed recent changes in tools and approaches to measuring urban area and urban

populations. While there have been rapid improvements in the availability and quality of global maps of

urban areas and population data derived from remote sensing sources, there remains no single or universal

map or population tool for documenting urban population growth and physical expansion. Current tools

available combining satellite maps with gridded population estimates are most advanced in this regard, but

still suffer from drawbacks in the availability and comparability of data inputs at the global level. There

are tradeoffs in terms of image resolutions, and the availability of comparable images over time. Methods

for determining the types of land use using daytime satellite imagery covers require careful calibration from

multiple images to both correct for surface interference and to accurately distinguish the colors and patterns

associated with various urban built environments across the world.

Nighttime lights data as well as the higher spectrum, non-optical satellite data products (such as the

Global Urban Footprint’s radar imagery) are an alternative option that does not require the same type of

corrections. These factors make broad estimates difficult as different types of urban land uses may or may

not be included in the urban land areas of different cities. Users also have different needs in terms of scale

and resolution; very high resolution data for example could be useful in examining urban fabrics of specific

cities, but other uses such as assessing economic activity or potential damages from natural disasters

utilizing corresponding GDP data could take advantage of coarser maps with other resolutions and

alternative frequencies (such as nighttime lights) more readily.

There has also been improvement in the reliability of population estimates for urban areas,

particularly the improved availability and granularity of census data for administrative units within

countries. This has been a key improvement for improving the accuracy of gridded population data sets.

Given the wide and arbitrary differences for defining urban places utilized by countries around the world,

Page 36: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

36

gridded population layers and standardized urban definitions have provided a way to comparably estimate

the number of people in urban areas across the world. They show that, contrary to WUP and administrative

area-bounded data, Africa, the Middle East and Asia have among the greatest shares of urban populations,

while Latin America comparatively less so. However, when utilizing the same population-based thresholds

for urban areas (such as minimum settlement population or density level) the estimates of urban populations

provided by different gridded population layers diverge. This is in part due to the different methods of

distributing population inputs within each grid, but further research is needed to better account for these

differences. These differences have important consequences for how governments can and should prioritize

investments to better accommodate new urbanization or leverage the advantages of existing cities.

Population-based urban estimation methods such as AI and Cluster also depend on defining the parameters

for measuring urban areas in addition to the gridded population layers used as inputs. If minimum

population and population density thresholds are low, more settlements will be classified as urban, though

there may be important qualitative differences between urban areas at the upper and lower bounds of this

distribution. This remains an area for continued exploration (Deuskar and Stewart 2016). Finally, there

currently exists no globally comparable panel set of built-up area maps linked to these two population-

based approaches. Such data would help to understand worldwide changes in the shape and location of

population clusters or agglomerations over time.

Studies in the remote sensing tradition - which equates urban areas with built-up areas or areas

which are brightly lit at night - show that urban expansion is occurring in different patterns across the world.

There also appears to be substantial evidence from available case studies that many cities (apart from urban

agglomerations in East Asia and selected countries such as Pakistan in South Asia) are becoming less dense

as land consumption rates eclipse population growth rates. In Sub Saharan Africa, Latin America and parts

of East Asia, discontinuous and leapfrog urban expansion appears to have become increasingly common in

recent decades, suggesting avenues for future research in identifying what factors may be driving this type

of expansion. Another commonality is the expansion of urban agglomerations across administrative

boundaries, both in regions where rural-urban migration or natural population increases are coinciding with

Page 37: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

37

urban expansion. These large and contiguous agglomerations, such as those in northern India and the Pearl

River Delta in China raise questions about how polycentric urban forms emerge and how current

governance arrangements can support and sustain them.

There also remain a number of topics for continued exploration and refinement utilizing satellite

mapping and gridded population data. For example, recent work by the World Bank has linked GDP data

with population and built up layers to estimate the value of potential damages under different natural

disaster scenarios. Additional refinement for global land cover mapping is needed, despite advancements

in algorithms and machine learning for classifying land use classes and types, there remain considerable

differences between the accuracy of these tools across spatial scales. This is due in part to differences in

how image analysts construct algorithms; the parameters they use and the inputs they are based on (Ban et

al. 2015). Location-based social media and geo-coded data could also be integrated into mapping exercises

(an example is GPW’s use of Facebook technology) and monitoring, especially as a means to confirm or

ground truth interpretations of objects or layers of built up areas, or to report and track public health issues

or disaster events “using humans as sensors” (Mertes et al. 2015, pg. 345).

Other avenues of research interest center on improving our understanding of urbanization and how

it relates to climate change impacts. This includes tracking the amount and location of arable land around

cities, the location and intensity of carbon emissions and assessing the risks of potential flooding from rising

sea levels. For example, some 50 percent of global urban land cover is within 116km of water bodies (Angel

2011; McGranahan et al. 2007). City level mapping exercises distinguishing urban land use types and

densities as well as vacant land for urban regeneration and infill development potential. This can also

improve options for restorative ecologies that can improve water collection and drainage by mapping

impermeable surfaces and the potential for urban agriculture.

Finally, apart from Huang et al. (2007) and Sevstuk and Amindarbari (2012) there has been little

comparative work classifying differences in the specific morphological dimensions of different urban

forms, such as shapes, levels of centrality, density and differences in shares of open space within built up

Page 38: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

38

areas. This would provide deeper insight on correlates between different types of urban morphologies and

other variables at the city or metropolitan level, such as, the location and extent of informal settlements,

access to services, economic activity and mobility, among others (Chakraborty et al. 2015). Such research

could begin to develop a typology of dimensions of urbanization which can in turn be used for calibrating

comparative land cover and urban population analyses. This would also have particular relevance to case-

specific analyses of shrinking cities, where there is a paucity of research to date utilizing remote sensing

tools to document or compare changes in declining urban populations.

Page 39: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

39

References:

Abrahams, A., Lozano-Gracia, N. and C.J. Oram. 2016. “Deblurring nighttime lights” Working Paper

Alonso, W. 1964. Location and Land Use Harvard University Press: Cambridge MA

Angel, S. Sheppard, S.C. and Daniel L. Civco. 2005. The Dynamics of Global Urban Expansion The

World Bank: Washington DC

Angel, S. Parent, J., Civco, D. and Alejandro M. Blei. 2010. “The persistent decline in urban densities:

Global and historical evidence of ‘sprawl’” Lincoln Institute of Land Policy Working Paper.

Angel, S., Parent, J., Civco, D., Blei, A. and David Potere. 2011. “The dimensions of global urban

expansion: estimates and projections for all countries, 2000-2050” Progress in Planning Vol. 75,

No.2, pp 53-107.

Atzberger, C. 2013. “Advances in remote sensing of agriculture: Context description, existing operational

monitoring systems and major information needs” Remote Sensing Vol. 5, No. 2. Pp. 949-981,

Bagan, H., and Yamagata, Y. 2015. “Analysis of urban growth and estimating population density using

satellite images of nighttime lights and land-use and population” GIScience & Remote Sensing Vol.

52, No.6, pp. 765-780.

Ban, Y., Gong, P., and Chandra Giri. 2015. “Editorial: Global land cover mapping using Earth

observation satellite data: Recent progresses and challenges>” ISPRS Journal of Photogrammetry

and Remote Sensing Vol. 103, pp.1-6.

Baruah, N. 2015. “Splintered and segmented? Fragmentation of African cities’ footprints” Draft

presentation for Spatial Development of African Cities Workshop. World Bank December 16-17.

Batty, M. 2008. “The size, scale and shape of cities” Science Vol. 319, No. 5864. pp. 769-771

Baugh, K., Elvidge, C., Zhizhin, M. and Feng Chi Hsu. 2014. “Using VIIRS NDB SDRs to Generate

Nighttime Lights Composites” National Geophysical Data Center

http://www.star.nesdis.noaa.gov/star/documents/meetings/2014JPSSAnnual/dayTwo/07_Session4a_

Baugh_DNB_20140513.pdf

Bertaud, A. and Malpezzi, S. 2014. “The spatial distribution of population in 57 World Cities: The role of

markets, planning and topography” Working Paper

Bertaud, A.2002. “The spatial organization of cities: Deliberate outcome or unforeseen consequence”

WDR Background paper 27864.

Bagan, H. and Yamagata, Y. 2014 “Land-cover change analysis in 50 global cities by using a

combination of Landsat data and analysis of grid cells” Environmental Research Letters Vol. 9, No.

6

Balk, D. 2009. “More than a name: Why is Global Urban Population Mapping a GRUMPy proposition?”

in Global Mapping of Human Settlement: Experiences, Data Sets and Prospects, edited by P. Gamba

and M. Herold, pp. 145-161. New York: Taylor and Francis

Balk, D., Pozzi, F., Yetman, G., Deichmann, U., and A Nelson. 2005. “The distribution of people and the

dimension of place: Methodologies to improve the global estimation of urban extents” in

Proceedings of the Urban Remote Sensing Conference of the International Society for

Page 40: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

40

Photogrammetry and Remote Sensing Paris: International Union for the Scientific Study of

Population.

Bloom, D.E., Canning, D., Fink, G., Khanna, T. and Patrick Salyer. 2007. “Urban settlement: data,

measures, and trends” Harvard School of Public Health Program on the Global Demography of

Aging, Working Paper Series.

Bloom, D. E., Canning, D. and Gunther Fink. 2008. “Urbanization and the wealth of nations” Science Vol

319, pp. 772-775.

Bundervoet, T. Maiyo, L. and Apurva Sanghi. 2015. “Bright lights, big cities: Measuring national and

subnational economic growth in Africa from outer space, with an application to Kenya and

Rwanda.” World Bank Policy Research Working Paper 7461

Chant, S. 2013. “Cities through a “gender lens:” A golden “urban age” for women in the global south?”

Environment & Urbanization Vol. 25, No.1, pp. 9-29.

Charkraborty, A., Wilson, B., Sarraf, S. and A. Jana. 2015. “Open data for informal settlements: Toward a

user’s guide for urban managers and planners” Journal of Urban Management Vol. 4., No. 2. Pp 74-

91.

Chomitz, K. M., Buys, P., and Thomas, T. S. 2005. “Quantifying the rural-urban gradient in Latin

America and the Caribbean.” World Bank, Development Research Group, Infrastructure and

Environment Team. Washington, DC: World Bank.

Clark, C. 1951. “Urban population densities” Journal of the Royal Statistical Society Vol 114, No.4, pp

490-496.

Cohen, B. 2004. “Urban growth in developing countries: A review of current trends and a caution

regarding existing forecasts” World Development Vol. 32, No. 1, pp. 23-51.

Deng, X., Hunag, J., Rozelle, S. and Emi Uchida. 2008. “Growth, population and industrialization, and

urban land expansion of China” Journal of Urban Economics Vol 63, No.1 pp 96-115.

Deichmann, U., Balk, D. and Yetman, G. 2001. “Transforming Population Data for Interdisciplinary

Usages: From census to grid”

Deuskar, C., and Stewart B. 2016. “Measuring global urbanization using a standard definition of urban

areas: Analysis of preliminary results” Draft Report No. AUS8082: World Bank

Dijkstra, L. and Poelman, H. 2014. “A harmonized definition of cities and rural areas: The new degree of

urbanization” European Commission Regional Working Paper WP 01/2014

Doll, C. 2008. “CIESIN thematic guide to night-time lights remote sensing and its application”

Socioeconomic Data and Application Center (SEDAC)

Doll, C.N. H., and Pachuari, S. 2010. “Estimating rural populations without access to electricity in

developing countries through nighttime satellite imagery” Energy Policy Vol. 38, pp. 5661-5670.

Doxey-Whitfield, MacManus, K., Adamo, S.B., Pistolesi, L., Squires J., Borkovska, O. and Sandra R

Baptista. 2015. “Taking advantage of the improved ability of census data: A first look at the Gridded

Population of the World, Version 4” Papers in Applied Geography Vol. 1, No. 3, pp. 226-234.

Page 41: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

41

Elvidge, C.D., Baugh, K.E., Kihn, E.A., Kroehl, H.W., and E.R. Davis. 1997. “Relation between satellite

observed visible-near infrared emissions, population, and energy consumption.” International

Journal of Remote Sensing Vol. 18, 1373-1379.

Elvidge, C.D., Sutton, P.C., Ghosh, T., Tuttle, B.T., Baugh, K.E., Bhaduri, B., and E. Bright. “A global

poverty map derived from satellite data” Journal of Computers & Geosciences Vol. 35, No.8, pp.

1652-1660.

Fay, M., and Opal, C. 2000. “Urbanization without growth: a not-so-uncommon phenomenon.” World

Bank Policy Research Working Paper 2412.

Fragkias, M., Lobo, J., and K.C. Seto. 2016. “A comparison of nighttime lights data for urban energy

research: Insights from scaling analysis in the US system of cities” Environment and Planning B:

Planning and Design Vol.

Glaeser, E. and Mare, D. 2001. “Cities and skills” Journal of Labor Economics Vol. 19, No. 2, pp 316-

342.

Glaeser, E. 2011. Triumph of the city: How our greatest invention makes us richer, smarter, greener,

healthier and happier. Penguin: New York

Graesser, J., Cheriyadat, A., Vatsavai, R.R., Chandola, V., Long, J., and Eddie Bright. 2012. “Image

based characterization of formal and informal neighborhoods in an urban landscape? IEEE Journal

of Selected Topics in Applied Earth Observations and Remote Sensing Vol.5, No.4, pp1164-1176.

Gunakesera, R., Ishizawa, O., Aubrecht, C., Blankespoor, B., Murray, S., Pomonis, A., and James Daniell.

2015. “Developing an adaptive global exposure model to support the generation of country disaster

risk profiles” Earth Science Reviews Vol. 150, pp. 594-608.

Hagan, J.M., 1998. “Social networks, gender, and immigrant incorporation: Resources and constraints”

America Sociological Review Vol.63. No.1, pp 55-67.

Hay, S.I., Noor, A.M., Nelson, A. and A.J. Tatem. 2005. “The accuracy of human population maps for

public health application” Tropical Medicine and International Health Vol. 10, No. 10, pp. 1073-1086

Hammer, S. Kamal-Chaoui, Robert, A. and Marissa Plouin. 2011. “Cities and green growth: A conceptual

framework” OECD Regional Development Working Papers 2011/08. OECD.

Henderson, Vernon, Adam Storeygard and David N. Weil. 2012. “Measuring economic growth from outer

space,” American Economic Review, 102(2): 994-1028.

Henderson, J.V., Venables, A.J., Regan, T., and Ilia Samsonov. 2016. “Building functional cities” Science

Vol. 352, No. 6288, pp 946-947.

Henderson, J. V., Regan, T., and Anthony J. Venables. 2016b. Building the city: sunk capital, sequencing,

and institutional frictions. CEPR Working Paper.

Henderson, Vernon and Nigmatulina, Dzhamilya. 2016c. The fabric of African cities: How to think about

density and land use. Draft April 20th, 2016. The London School of Economics.

Hoffmann, P., Strobl, J., Blaschke, T., and H. Kux. 2008. Detecting informal settlements from QuickBird

data in Rio de Janeiro using an object based approach. In T. Blaschke, S. Lang & G. Hay (Eds.),

Object-Based Image Analysis (pp. 531-553): Springer Berlin Heidelberg.

Page 42: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

42

Huang, J., Lu, X.X., and J. Sellers. 2007. “A global comparative analysis of urban form: Applying spatial

metrics and remote sensing” Landscape and Urban Planning Vol. 82, pp. 184-197.

IIHS (Indian Institute for Human Settlements). 2011. Urban India 2011: Evidence. Bangalore: IIHS.

Inostroza, L. Baur, R. and Elmar Csaplovics. 2013. “Urban sprawl and fragmentation in Latin America: A

dynamic quantification and characterization of spatial patterns” Journal of Environmental

Management Vol. 115, pp. 87-97.

Ionkova, K. and Sulukhia, T. 2015. Ukraine Urbanization Review World Bank: Washington DC

Jacobs, J. 1969. The Economy of Cities New York: Random House.

Karim, M. S., and A. Nasar. 2003. “Migration Patterns and Differentials in Pakistan: Based on the

Analysis of the 1998 Census Data.” In Population of Pakistan: An Analysis of 1998 Population and

Housing Census, edited by A. R. Kemal, M. Irfan, and N. Mahmood,173–80. Pakistan Institute of

Development Economics: Islamabad.

Kim, A. M., Gong, P., and Liu, D. 2004. “Change Detection from SPOT-Panchromatic Imagery at the

Urban-Rural Fringe of Ho Chi Minh City, Vietnam” Geographic Information Sciences, Vol. 10, No.

1, pp.:42-48.

Klotz, M., Kemper, T., Geiß, C., Esch, T., and Taubenböck, H. 2016. Mapping spatial settlement patterns

on a global scale: Multi-scale cross-comparison of new and existing global urban maps.

Kourtit, K., Nijkamp, P., Franklin, R.S. and Andres Rodriguez Pose. 2014. “A blueprint for strategic

urban research: the urban piazza” Town Planning Review Vol. 85, No.1, pp. 97-126.

Kuffer, M., Pfeffer, K., and Richard Sliuzas. 2016. “Slums from space: 15 years of slum mapping using

remote sensing” Remote Sensing Vol. 8, No.6, pp. 455

Landscan. 2016. “Landscan documentation”

http://web.ornl.gov/sci/landscan/landscan_documentation.shtml#01 Accessed Sept 7 2016.

Libertun de Duren, N. and Compean, Roberto. 2015. “Growing resources for growing cities: Density and

the cost of municipal public services in Brazil, Chile, Ecuador and Mexico” IDB Working Papers

IDB-WP-634

Li, B. 2006. “Floating population or urban citizens? Status, social provision and the circumstances of rural-

urban migrants in China” Social Policy and Administration Vol. 40, No.2, pp. 174-195.

Linard, C. Gilbert, M. and A.J. Tatem. 2011. “Assessing the use of global land cover data for large area

population distribution modeling” GeoJournal Vol. 76, No.5, pp. 525-538.

Lozano-Gracia, N. Antos, S.E., Jimenez, R., and A.I. Aguilera. 2016. “Morphology of cities in Central

America” Manuscript.

Maria, A., Acero, J.L. and A. Aguilera Making Cities Work for Central America Urbanization Review

Washington, DC: World Bank

Marshall, A. 1961. Principles of Economics 9th edition. Macmillan: London

Page 43: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

43

McGranahan, G., Balk, D., and B. Anderson. 2007. “The rising tide: Assessing the risks of climate change

and human settlements in low elevation coastal zones” Environment and Urbanization Vol. 19, No. 1,

pp. 17-37.

McGranahan, G., Schensul, D., and G. Singh. 2016. “Inclusive urbanization: can the 2030 Agenda be

delivered without it?” Environment & Urbanization Vol. 28., No.1, pp. 13-34.

McNeil, J.R. 2000. Something new under the sun: An environmental history of the twentieth century. New

York: W. W. Norton.

Molina, G.G., Perez Rada, E., and W. Jimenez. 2002. “Social exclusion: Residential segregation in Bolivian

cities” Inter-American Development Bank Research Networking Paper #R-440.

McMillen, D.P. and Smith, S.C. 2003. “The number of subcenters in large urban areas” Journal of Urban

Economics Vol. 53, No.2, pp. 321-338.

Mellander, C., Stolarick, K., Matheson, Z., and Jose Lobo. 2013. “Night-time light data: A good proxy

measure for economic activity?” Martin Prosperity Research Working Paper MPIWP 006.

Mertes, C. M., Schneider, A., Sulla-Menashe, D., Tatem, A.J. and Tan, B. 2015. “Detecting change in

urban areas at continental scales with MODIS data.” Remote Sensing of Environment Vol. 158, pp.

331-347.

Montgomery, M.R. 2008. “The urban transformation of the developing world” Science Vol. 319. No.

5864, pp 761-764.

Montgomery, M.R. and D. Balk, 2011. “The Urban Transition in Developing Countries: Demography

Meets Geography” in E. Birch and S. Wachter eds., Global Urbanization. Philadelphia: University

of Pennsylvania Press.

Muth, R.F. 1969. Cities and housing: The spatial pattern of urban residential land use University of

Chicago Press: Chicago.

Pallagst., K., Aber, J., Audirac, I., Cunningham-Sabot, E., Fol, S., Martinez-Fernandez, C., Moraes, S.,

Mulligan, H., Vargas-Hernandez, J., Wiechmann, T., Wu., T. and Jessica Rich. 2009. The Future of

Shrinking Cities: Problems, Patterns and Strategies of Urban Transformation in a Global Context

Institute of Urban and Regional Development (IURD): Berkeley, CA

Pesaresi, M., Huadong, G., Blaes, X., Erlich, D., Gueguen, L., Halkia, M., Kauffman, M., Kemper, T.,

Lu, L., Marin-Herrra, M.A., Ouzounis, G.K., Scavazzon, M., Soille, P., Syrris, V., and Luigi

Zanchetta. 2013. “A Global Human Settlement Layer from optical HR/VHR RS data: Concept and

first results.” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Vol. No. pp. 2102-2131

Potere, D., Schneider, A., Angle, S. and Daniel L. Civco. 2009. “Mapping urban areas on a global scale:

which of the eight maps is more accurate?” International Journal of Remote Sensing Vol. 30, No.

24, pp. 6531-6558.

Richardson, H.W., Chang-Hee, C.B., and M. Baxamusa. 2000. “Compact cities in developing countries:

Assessment and implications” in Compact cities: Sustainable urban forms for developing countries

Mike Jenks and Rad Burgess, eds., Taylor and Francis: New York.

Page 44: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

44

Roberts, M., Blankespoor, B., Deuskar, C., and Stewart, Benjamin. 2016. “Urbanization and

Development: Is Latin America and the Caribbean different from the rest of the world?”,

Background Paper to the LAC Cities and Productivity Flagship Report, World Bank: Washington,

D.C.

Satterthwaite, D. 2007. The transition to a predominantly urban world and its underpinnings. London:

Institute for Environment and Development (IIED).

Satterthwaite, D. 2010. “Urban myths and the mis-use of data that underpin them” UNU-WIDER

Working Paper No. 2010/28

Schneider, A. and Woodcock, C.E. 2008. “Compact, dispersed, fragmented, extensive? A comparsion of

urban growth in twenty-five global cities using remotely sensed data, pattern metrics and census

information” Urban Studies Vol. 45, No.3, pp. 659-692

Schneider, A., M A Friedl and D Potere 2009. “A new map of global urban extent from MODIS satellite

data” Environmental Research Letters, Vol. 4, No. 4, pp1-11.

Schneider, A. and Mertes, C.M. 2014. “Expansion and growth in Chinese cities, 1978–2010”

Environmental Research Letters Vol. 9, No. 2, pp. 1-11.

Schnieder, A. Mertes, C.M., Tatem, A.J., Tan, B. Sulla-Menashe, Graves, S.J., Patel, N.N., Horton, J.A.,

Gaughan, A.E., and J.T. Rollo. 2015. “A new urban landscape in East–Southeast Asia, 2000–

2010“Environmental Research Letters Vol. 10, No. 3

Scott, A.J. 2001. Global City-Regions: Trends, Theory, Policy Oxford University Press: Oxford, UK

Seto, K.C., Fragkias, M., Guneralp, B., and Michael K. Kelly. 2011. “A meta-analysis of global urban

land expansion” PLoS ONE Vol. 6, No. 8., pp. 1-9.

Seto, K.C., Sanchez-Rodriguez, R., and Michail Fragkias. 2010. “The new geography of contemporary

urbanization and the environment” Annual Review of Environment and Resources Vol. 35, No.1, pp.

167-194.

Sevstuk, A. and Amindarbari, R. 2012. “Measuring growth and change in East-Asian cities: Progress

report on urban form and land use measures”

Shepard, W. 2015. Ghost cities of China: The story of cities without people in the world’s most populated

country Zed Books: London.

Shi, K. Yu, B. Huang, Y., Hu, Y., Yin, B. Chen, Z. Chen, L and Jianping Wu. “Evaulating the Ability of

NPP-VIIRS Nighttime Light Data to Estimate Gross Domestic Product and Electric Power

Consumption of China at Multiple Scales: A Comparison with DMSP-OLS Data” Remote Sensing

Vol. 6, pp. 1705-1724.

Socioeconomic Data and Applications Center (SEDAC). 2016. “Global Rural-Urban Mapping Project

(GRUMP), v1” http://sedac.ciesin.columbia.edu/data/collection/grump-v1

Small, C., and Elvidge, C.D. 2012. “Night on Earth: Mapping decadal changes of anthropogenic night

light in Asia” International Journal of Applied Earth Observation and Geoinformation Vol. 22, pp.

40-52.

Page 45: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

45

Stevens, F. Gaughan, A., Linard, C., and A. Tatem. 2015. “Disaggregating census data for population

mapping using random forests with remotely-sensed and ancillary data. PloS ONE Vol. 10, No. 2.

Storper, M. and Venables, A. 2004. “Buzz: face to face contact and the urban economy” Journal of

Economic Geography Vol.4, No.4, pp 351-370.

Sudjic, D. and R. Burdett, 2007. The Endless City Phaidon: London

Tatem, A.J., Noor, A.M., and Simon I Hay. 2004. “Defining approaches to settlement mapping for public

health management in Kenya using medium spatial resolution satellite imagery” Remote Sensing and

the Environment Vol. 93, Nos. 1-2, pp.42-52.

Taubenböck, H., Roth, A., Esch, T., Felbier, A., Müller, A., & Dech, S. 2011. “The vision of mapping the

global urban footprint using the TerraSAR-X and TanDEM-X mission.” Urban and regional data

management, 243-251.

Taubenbock, H., Wiesner, M., Felbier, A., Marconcini, M., Esch, T. and S., Dech. 2014. New dimensions

of urban landscapes: The spatio-temporal evolution from a polynuclei area to a mega-region based

on remote sensing data” Applied Geography Vol. 47, pp.137-153

Tiwari, M. Godfrey, N. et al. 2016. Better cities, better growth: India’s urban opportunity. New Climate

Economy, World Resources Institute and the Indian Council for Research on International Economic

Relations: London, Washington, D.C. and New Delhi.

Uchida, H. and Nelson, A. 2008. “Agglomeration Index: Towards a new measurement of urban

concentration.” World Development Report Background Paper

United Nations. 2008. World Urbanization Prospects: The 2007 Revision Department of Social and

Economic Affairs, United Nations: New York.

United Nations. 2014. World Urbanization Prospects: The 2014 Revision Department of Social and

Economic Affairs, United Nations: New York.

United Nations. 2016. “Goal 11: Make cities inclusive, safe, resilient and sustainable”

http://www.un.org/sustainabledevelopment/cities/ Accessed 30 September 2016.

United States Census Bureau. 2008. American Community Survey (ACS) Census Bureau: Washington

DC.

Venables, A. 2011. “Productivity in cities: self-selection and sorting” Journal of Economic Geography

Vol. 11. No. 2, pp. 241-251.

Vithayathil, T., and G. Singh 2012. “Spaces of discrimination: Residential segregation in Indian cities”

Economic & Political Weekly Vol. 47., No. 37, pp. 60-66.

Wackernagel, M., Kitzes, J., Moran, J., Goldfinger, S., and M. Thomas. 2006. “The ecological footpring

of cities and regions: Comparing resource availability with resource demand” Environment &

Urbanization Vol. 18., No.1, pp. 103-112.

World Bank. 2009. World Development Report: Reshaping economic geography World Bank:

Washington, DC.

Page 46: Bright Lights, Big Cities? Review of research and findings ...pubdocs.worldbank.org/en/720741492724279464/Bright... · provides an overview of these techniques as well as the main

46

World Bank. 2013. Inclusion Matters: The Foundation for Shared Prosperity World Bank: Washington,

DC

World Bank. 2013b. Planning, Connecting and Financing Cities – Now: Priorities for City Leaders

World Bank: Washington, DC.

World Bank. 2015. East Asia’s Changing Urban Landscape: Measuring a Decade of Spatial Growth

World Bank: Washington, DC

World Bank. 2016. “Opening doors to the world: Africa’s Urbanization” Draft

World Bank and Development Research Center of China (DRC) 2016. Urban China: Toward Efficient,

Inclusive and Sustainable Urbanization Washington DC: World Bank

Zhang, K., and Song, S. 2003. “Rural-urban migration and urbanization in China: Evidence from time-

series and cross-section analyses” China Economic Review Vol. 14, No.4. pp. 386-400.

Zhang, Q. and Seto, K.C. 2011. “Mapping urbanization dynamics at regional and global scales using

multi-temporal DMSP/OLS nighttime light data” Remote Sensing of Environment Vol. 115, pp.

2320-2329.

Zhou, Y., Smith, S.J., Zhao, K., Imhoff, M., Thomason, A. Bond-Lamberty, B., Asrar, G.R., Zhang, X.,

and Christopher Elvidge. 2015 “A global map of urban extent from nightlights” Environmental

Research Letters Vol. 10, No.5


Recommended