Abstract
This paper investigates the pressure placed by cities on their environment with respect to urban air pollution. The analysis employs a spatially explicit global dataset of emissions to estimate urban emissions of four pollutants from a sample of 8038 cities world-wide in 2005. A cross-sectional regression analysis is then conducted to examine the association of urban air pollution with socioeconomic and geographical factors. The results confirm that urban pollution is associated primarily, but not exclusively, with demographics. The results suggest that urban pollution is likely to increase with population growth and that economic modernisation is unlikely to provide much relief from the pressures placed by coming population growth. The findings suggest that policy-makers must focus on reducing the emissions intensity of production activities within cities, especially from the energy sector, if they are to avoid rapid growth in urban air pollution in coming decades.
Introduction
Cities have rapidly become the dominant organisational form of human society. In 1900, only 13 per cent of the global population lived in cities; by 2050, two out of three persons world-wide are expected to live in cities (United Nations Population Division, 2010). With this structural transition comes heightened concern over the environmental and human health impacts of urban development. For instance, many of the world’s largest cities suffer from severe air pollution and human health impacts that impair quality of life and shorten life expectancies (Chow et al., 2004; Gurjar et al., 2008). Some urban air pollutants, such as ozone, methane and particulate matter, are also evident as potent climate forcers at multiple scales (Akimoto, 2003; Horowitz, 2006). The global sustainability challenge will be to accommodate widespread urbanisation while improving living standards, maintaining healthy local environments and minimising the aggregate environmental impacts of development (MacNeil, 1990; Seto and Satterthwaite, 2010).
Policy-makers look to the research community for guidance on the best ways to leverage urban development to realise sustainable development goals. Nevertheless, the research literature is divided over the impact of cities on environmental degradation and in their potential to address environmental challenges. The divisions arise, in large part, due to the complex interrelationships of economic development with industrialisation, technological modernisation, population growth and urbanisation.
This paper investigates the pressure placed by cities on their environment with respect to urban air pollution. The analysis utilises global, spatially explicit datasets to estimate the air pollution released by cities in 2005. A regression analysis of city-level pollution is then conducted to examine the correlates of pollution. The results confirm that urban pollution is associated primarily, but not exclusively, with demographics. The results suggest that urban pollution is likely to increase with population growth and that economic modernisation is unlikely to provide much relief from the pressures placed by coming population growth. The goal of this paper is to provide an empirical base from which to consider further policy and pollution management efforts that may be required to combat air quality problems in the coming decades.
The Drivers of Environmental Degradation
The literature on environmental degradation is often based upon the theoretical construct of the ‘social metabolism’, which applies the biological metabolism metaphor to society and its environment (for example, Fischer-Kowalski and Hüttler, 1998). Under the metabolic framework, society takes in air and water inputs, consumes energy resources and generates waste by-products such as air and water pollution.
One hypothesis stemming from this metabolic framework is that larger societies will have larger energy demands, which will deplete resources faster and generate larger waste streams in the process than smaller societies. Indeed, early theorists focused on the role of population as a driver of environmental degradation, predicting that population growth would eventually outstrip the capacity of the environment to provide sustaining resources and services (such as food), which would lead to population declines (Malthus, 1798).
Contemporary scholars acknowledge that economic development and technological modernisation, among other factors, can mediate the impact of population growth on the environment. For instance, industrialisation and the ‘growth imperative’ accelerated the social metabolism, requiring increasingly intensive forms of energy to sustain economic growth resulting in widespread resource depletion and the production of large waste streams that taxed the absorptive and cleansing capacity of the natural environment (Gould et al., 2008; MacNeil, 1990). Globalisation and urbanisation further intensified and expanded the environmental impacts of economic development, as advanced societies could obtain energy resources and products for consumption from far distances, shifting environmental burdens away from their immediate environment to less developed regions and causing a ‘metabolic rift’ between natural and social systems (Foster, 2009; McGranahan, 2007). Scholars debate whether the future will follow historical patterns of increased environmental degradation, or could ‘decouple’ environmental impact from population and economic growth through the imposition of more efficient technologies and/or the emergence of stronger environmental sensitivities.
Given these complex interactions, the IPAT equation was proposed as a framework for envisioning how humans impact the environment, specified as I = PAT, where environmental impacts (I) are the product of population size (P), affluence (A) and technology (T) (Holdren and Ehrlich, 1974). In this case, technology broadly represents the efficiency of production and consumption activities with respect to environmental resources. Some scholars prefer to disaggregate the technology dimension of IPAT into a consumption dimension C (i.e. consumption of some good or service per unit GDP) and a technology impact dimension T (i.e. impact per unit consumption of some good or service), such as in the ImPACT or I = PACT specification (Waggoner and Ausubel, 2002). The interaction of population and affluence in the ImPACT model comprise the ‘sustainability challenge’, as described earlier, while the interaction of consumption and technology comprise the ‘sustainability levers’.
The IPAT framework has been frequently criticised for being oversimplistic regarding the key processes that may be responsible for generating environmental impact (Chertow, 2000; York et al., 2003b).
In addition, the underlying relationships between environmental impact and its key drivers (population, affluence and technology) may be subject to non-linearities and threshold effects. For instance, many scholars assume a proportional relationship between impact and population, meaning that impact would change by one unit with every one unit change in population, all else equal. Some scholars have questioned the appropriateness of this proportionality assumption. They ask: what if the environmental effect changes with population size? By including population size as its own factor, scholars can statistically test the type of scaling effect (Dietz and Rosa, 1994; York, 2007). The empirical evidence on scaling in relation to environmental degradation is mixed, with some evidence to suggest that pollution impact may decline with population (Fragkias and Lobo, 2010; York et al., 2003b) and other evidence to suggest that energy consumption and associated pollution increases with population (Bettencourt et al., 2007; Jorgenson and Clark, 2010; Parikh and Shukla, 1995; York, 2007). The findings vary depending on the environmental outcome studied, the units of analysis and the model specification used in the analysis.
As with population, some scholars have questioned whether the environmental impact of affluence may vary depending on its level. An inverted U-shaped relationship has been found between affluence and some measures of environmental impact, such as air and water quality and deforestation (Ehrhardt-Martinez et al., 2002; Ekins, 1997; Grossman and Krueger, 1995). This relationship has been termed the environmental Kuznets curve (EKC), drawing on a similar relationship found by Simon Kuznets between economic development and income inequality (Grossman and Krueger, 1995). Ecological modernisation theorists explain the EKC by noting that affluent societies may have the resources, institutions and desire to curb their environmental impact (for example, Mol, 2002). Such actions may not have been affordable or desirable to less-wealthy societies.
Empirical study of the EKC has generated mixed and contentious results that are sensitive to the model specification and outcome variable of interest (Dasgupta et al., 2002; Ekins, 1997). Some of the disagreement over the EKC may relate to the scale of health risks posed by the variable of interest. The urban environmental transition (UET) framework posits that community-scale risks (such as urban air pollution) could increase and then decrease with increasing wealth, while global-scale risks (such as from greenhouse gases) could rise steadily with increasing wealth (McGranahan, 2007; Wilkinson et al., 2007).
The logic of collective action provides one rationale for divergent society–environment relationships at different scales; small-area risks affect relatively few people, who may be easier to organise in response to the risk, while large-area risks affect many more people, who may be difficult to organise even when it is in their best interests to act (Olson, 1971). Indeed, Torras and Boyce (1998) argued that the ‘decoupling’ of impact from economic development seen in the EKC will not happen automatically as incomes rise, but rather depends on active citizen engagement and subsequent policy response to protect the environment. Thus, several researchers include measures of democratic governance or policy activity in their investigations of environmental degradation (Ehrhardt-Martinez et al., 2002; York et al., 2003a).
Finally, the literature is divided over the environmental impact of urban development. Some scholars include measures of urbanisation—i.e. the share of a country’s population living in cities—as a proxy for technological modernisation, where higher urbanisation rates are presumably associated with advanced technologies and institutions of modern society. These analyses have found a significant positive effect of urbanisation on national-level outcomes, including energy use, carbon dioxide emissions, ecological footprints and water pollution (Cole and Neumayer, 2004; Jorgenson and Clark, 2011; Parikh and Shukla, 1995; Torras and Boyce, 1998; York, 2007; York et al., 2003b). The results are consistent with the idea that the complexity of organisation within cities speeds up the social metabolism, resulting in high resource use and production of waste materials (Wolman, 1965).
Yet, the underlying processes generating environmental impact from urban development are not at all clear with such a crude measure as urbanisation rate. Urban scholars argue that “different forms of urbanization have different impacts on the local and global environment” (Seto et al., 2010, p. 169). The urbanisation–environment relationship may vary by development status, where urbanisation in the least developed nations may result in reductions in energy use opposite to what is seen for more developed nations, and which could be attributed to the unique metabolic dynamics found in urban slums (Jorgenson et al., 2010; Poumanyvong and Kaneko, 2010). Experiences within developed nations suggest that sprawling forms of urban development may have larger aggregate impacts on the environment than compact-style development, owing in part to powerful business interests that gain from extensive low-density and energy-inefficient construction (Gonzalez, 2005).
A more nuanced approach using sub-national data appears necessary to investigate the activities and processes at work within cities that may influence the environment. Previous studies examining city-level data often focus on the environmental impact of the world’s largest cities, such as ‘megacities’ with more than 10 million residents (Butler et al., 2008; Gurjar et al., 2008; Guttikunda et al., 2005; He et al., 2002; Sovacool and Brown, 2010). Other studies focus mostly on cities in developed countries that are more likely to have available data than their developing counterparts (Lobo et al., 2009; Scholz, 2006). Our knowledge of emissions activity in smaller cities and from developing cities is limited, especially in a comparative global context. Such a gap in knowledge is problematic because almost all of the world’s future population growth is expected to occur in moderate-sized cities in developing regions (United Nations Population Division, 2010). Thus, there remains a need to reach a better understanding of the global geography of urban air pollution, which could assist in evaluating emission mitigation potential from cities and in assessing the effectiveness of emissions reduction activities within cities and their environs. Extensive data limitations have so far prevented such an endeavour.
Methods and Data
The primary objective of this analysis is to explore the spatial pattern of urban air pollution world-wide and the primary correlates of urban air pollution in a comparative context. The analysis employs a spatially explicit global dataset of emissions to estimate separately the urban emissions of four pollutants from a sample of 8038 cities world-wide in 2005. The analysis then uses a cross-sectional regression model to examine the relative influence of population size, density, economic development, industrial composition, growth rate and climate on emissions. The model specification was designed to investigate several issues previously raised, such as population and income scaling effects.
Conceptual Model
Dietz and Rosa (1994) developed the STIRPAT model for empirically testing hypotheses related to IPAT, denoted as
The regression coefficients b, c and d can be interpreted as elasticities—i.e. the percentage change in impact associated with a 1 per cent change in the independent variable. The coefficient b in equation (1) can also be used to test population scaling effects. Coefficients for b that are not significantly different from 1 would support the proportionality assumption; coefficients that are significantly smaller than 1 indicate that marginal impact declines with increases in population; while coefficients that are significantly larger than 1 indicate that marginal impact increases with population size.
This analysis employs a modified STIRPAT specification that builds on previous literature and is tailored to cities and to available datasets. One important challenge in conducting a city-level analysis is in obtaining meaningful data to represent each independent variable at the sub-national level, especially for technology. Many indicators are only available at the national level or for a small subset of large cities. The model employed here is
where, E represents urban emissions; P is urban population; A is GDP per capita; D is population density;
Impact Variables
This paper separately examines the spatial production of four air pollutants in 2005: nitrogen oxides (NOX), non-methane volatile organic compounds (VOCs), carbon monoxide (CO) and sulphur dioxide (SO2). NOX, VOCs and CO contribute to the formation of surface-level ozone (O3), which is a respiratory irritant and can be toxic at high concentrations. SO2 and NOX react with water vapour to form acids, which harm vegetation and degrade building surfaces and react with other substances to form particulate matter (PM). These four substances, in combination with O3 and PM, pose significant human health risks in cities and remain a focus of pollution regulation world-wide (Akimoto, 2003). Future research could apply this approach to other environmental hazards, such as greenhouse gases.
Pollution emissions data were obtained from the Emission Database for Global Atmospheric Research (EDGAR), version 4.1. The data are available by compound and by source category on a global grid (0.1 degree resolution). Emissions data were derived from national emissions inventories and spatially allocated according to location, where possible, such as for industrial and transport sources. Where spatial information was lacking, emissions were allocated by population using the Gridded Population of the World version 3 dataset produced by the Center for International Earth Science Information Network (CIESIN) at Columbia University. A few pollution source categories were not spatially gridded within EDGAR and are excluded, such as for forest fires and savanna burning. EDGAR also does not include biogenic emissions. Thus, the EDGAR dataset is modelled and does not represent directly observed emissions of polluting activities. Nevertheless, EDGAR represents one of the most comprehensive and spatially explicit emissions inventories available (Butler et al., 2008).
One challenge involved matching emissions data with city boundaries. Here, the analysis utilised urban boundaries and population data from the Global Rural–Urban Mapping Project (GRUMP) Beta version, produced by CIESIN. The GRUMP 1990 and 2000 population grids (at 30 arc-seconds resolution) were constructed to match the UN’s World Urbanisation Prospects (2003) share of urban population by country. The allocation of urban population within a country was based on night-time satellite imagery, among other sources. GRUMP identified discrete boundaries for cities (circa 2000) and linked the population data with administrative data so that the individual urban boundaries can be identified. Multiple contiguous cities were merged into larger urban agglomerations.
Emissions from grid cells were assigned to urban extents according to the share of the grid cell covered by the extent. The emissions assignment procedure captured only the emissions that were produced within the urban extent boundaries as identified by GRUMP. This procedure does not include emissions from activities outside city boundaries that may be closely linked to cities, such as agriculture, energy production and waste remediation. The generous urban extent boundaries from GRUMP are likely to include more peri-urban activity than might otherwise be captured with narrower urban boundary classifications (such as MODIS or Global Land Cover; see Schneider et al., 2009).
Independent Variables
Population (P)
Population data for 1990 and 2000 were derived from the GRUMP population grids. Each urban extent’s 2000 population and average annual population growth rate (1990–2000) were used to extrapolate its population to 2005. Extents with fewer than 50 000 residents in 2005 were excluded from the analyses. Urban extents with larger populations are expected to have higher emissions, all else equal.
GDP per capita (A)
This model employs GDP per capita as a measure of affluence. The GDP data were obtained from the International Institute for Applied Systems Analysis (IIASA), Greenhouse Gas Scenario Database. This analysis uses GDP per capita for the B1 scenario for 2000 and 2010 to interpolate mid-range values for 2005. 1 Urban extents with higher GDP per capita are expected to have higher emissions, all else equal.
The model includes mean-centred quadratic and cubic terms for GDP per capita. Positive and significant coefficients for GDP per capita combined with negative and significant coefficients for the quadratic term provide evidence of non-constant effects from affluence, as hypothesised by the EKC. Positive and significant coefficients on the cubic term indicate that the income–pollution relationship changes direction at the highest levels of affluence as posited by some critics of ecological modernisation.
Density (D)
This model employs a variable for population density, calculated as the total population within the city divided by the square kilometres of land area within the urban extent (after excluding area covered by water or ice using the Global Land Cover 2000 dataset). Urban extents with higher population densities are expected to have lower emissions, all else equal.
Share of emissions from energy production, industry, transport (
,
,
)
The regressions were run with variables for the share of urban emissions by pollutant from the energy sector, from the non-energy industrial sector and from road transport. The inclusion of these variables helps to account for otherwise anomalous emissions from urban areas with specialised functions. 2 In general, areas with concentrated production activities are expected to have higher emissions.
Growth rate (R)
The model includes a variable representing recent population growth. It is calculated as the annual average population growth rate, using GRUMP data for 1990 and 2000. The growth rates were limited to a maximum of 10 per cent annually and a minimum of -5 per cent annually, which reflect the outer fences of the interquartile range. The relationship between population growth rates and emissions is indeterminate a priori. Cities facing rapid population growth may have a hard time keeping up with the changing activity associated with rapid population growth and thus may have higher environmental degradation contemporaneously. Alternatively, rapidly-growing cities are more likely to be less developed and thus rapid growth may be associated with lower emissions.
Climate (C)
Ecologists and geographers highlight the role of climate and biophysical aspects of the environment on social organisation and economic activity, and thereby on environmental impact (Grimm et al., 2008; Pickett et al., 1997; York et al., 2003a). The model includes a variable that indicates the extent of cooling necessary to maintain comfortable living temperatures. Monthly temperature data used to construct annual cooling degree days (CDD) by location were obtained from the Climatic Research Unit at the University of East Anglia. The global temperature data were averaged over 30 years (1961–1990), reflecting longer-running climate patterns. The relationship between climate and emissions is also indeterminate a priori.
Development status
Previous literature indicates that the relevant relationships and effect sizes may be contingent on the economic development status of the study units (Poumanyvong and Kaneko, 2010). The regressions were run using the full sample of cities as well as segmented by three development categories offered by the United Nations Statistical Division: developed nations (North America, Europe, Japan, Australia and New Zealand); developing nations (Africa, Central and South America, Asia excluding Japan, and Oceania excluding Australia and New Zealand); and least developed nations (such as Afghanistan, Bangladesh and Ethiopia). The regression results were qualitatively similar for all samples and only the full sample results are presented here.
Results
Urban Air Pollution
According to this analysis, the world’s 8038 urban areas with more than 50 000 estimated residents produced approximately 40 per cent of the anthropogenic pollution of NOX, VOCs and SO2 and 32 per cent of CO pollution in 2005 (see Table 1). The sample urban areas contained approximately 87 per cent of the global urban population and approximately 90 per cent of the global urban NOX, VOCs, CO and SO2 pollution in 2005. If all urban areas with available emissions data are included, the urban shares of global population, NOX, VOCs and SO2 emissions would be approximately 50 per cent and the urban share of CO would be 35 per cent. As mentioned, these amounts represent the pollution directly produced within urban extent boundaries. These totals are not adjusted upwards for urban-related pollution released outside urban bounds (such as for energy production or waste management), or adjusted downwards for export-related pollution for products consumed elsewhere. Thus, the totals represent a best estimate of the ‘partial footprint’ of global urban pollution for these four pollutants in 2005.
Distribution of urban air pollution and population by region and development status
Emissions listed in metric tons of NO2.
Notes: The urban emissions reflect pollution for 8038 urban extents world-wide with more than 50 000 estimated residents in 2005. The numbers in parentheses reflect the urban share of global terrestrial emissions and population. Shaded cells have urban shares of more than 50 per cent. Region and development status from the United Nations Statistical Division.
The results affirm that pollution-generating economic activities tend to concentrate within large urban areas. Many of the world’s most populated urban agglomerations produced the most pollution in 2005, with Tokyo, Shenzhen, New York, Seoul, Jakarta, São Paulo, Los Angeles, Taipei and Shanghai all falling in the top 10 rankings for at least one pollutant (see Table 2). Several medium-sized urban areas also emerge in the top rankings according to this analysis, including Ulsan, Salmiya and Ufa.
Top 10 largest polluting urban areas
Notes: for cities with more than 50 000 estimated residents in 2005. The names used here follow GRUMP’s naming conventions. For instance, GRUMP’s ‘Tokyo’ urban extent boundary includes much of the ‘bullet train’ corridor of Tokyo–Nagoya–Kyoto–Osaka in Japan, with an estimated population of 78 million in 2005. GRUMP’s ‘New York’ boundary includes the New York–Northern New Jersey–Long Island and Philadelphia–Camden–Wilmington metropolitan areas, with a combined population of 27 million in 2005.
The concentration of pollution activity in some of the smaller and medium-sized urban areas changes the geography of the largest emitters when viewed on a per capita basis (see Table 2). Most of the largest per capita emitters are areas with fewer than 500 000 estimated residents in 2005. High on the per capita rankings are urban areas such Kozani, Traralgon, Magnitogorsk and Rouyn-Noranda, which have extensive energy-sector activities such as mining, petroleum refining and power generation.
Results derived using global datasets for any particular city are expected to deviate from findings that employ local or regional assessments (Guttikunda et al., 2005), owing to differences in urban boundaries, emissions sources evaluated and estimation procedures (Butler et al., 2008; Gurjar et al., 2008).
Regression Analysis
This analysis estimates the contemporaneous correlation of various socioeconomic and geographical factors with estimated air pollution among a global sample of urban areas in 2005. The regressions were estimated with ordinary least squares (OLS) and robust standard errors (see Table 3). The regressions explain at least three-quarters of the variation in estimated urban air pollution.
Estimation results
Natural log.
Higher-order terms for GDP per capita were mean-centred to avoid extreme collinearity.
Notes: Robust standard errors in parentheses. *** p < 0.01; ** p <0.05; * p < 0.1.
First, the results support the hypothesis that environmental impact is associated primarily, but not exclusively, with demographics (Grimm et al., 2008). Population size has the largest standardised effect on city-level pollution for all four pollutants. The analysis here also finds that population has a more than proportional impact on emissions in cities, with the largest effect sizes in cities from within the least developed nations. The results indicate ‘super-linear’ scaling from population size after controlling for other plausible factors (Bettencourt et al., 2007). The population scaling effect may result because larger cities are more complex, requiring more energy, which translates into the production of additional waste pollution under the metabolic framework (Wolman, 1965). One practical implication of this finding is that using per capita emissions as a dependent variable, which presumes a proportional effect from population on emissions, may obscure scaling effects at the margin and may confound the relative effects of population with other variables.
An examination of the bivariate relationship between population size and the predicted values for each city suggests stronger population scaling effects for smaller cities (i.e. steeper slope) than for larger ones. The effect is affirmed when segmenting the sample by population size classes, where the magnitude of the scaling coefficient is substantially larger for cities with fewer than 1 million residents than for larger cities. 3 Thus, if the cross-sectional findings accurately reflect an underlying structural relationship between population size and emissions, then an increase in population in smaller cities would be expected to result in a larger relative increase in pollution than an increase in population in larger cities, all else equal. This result is sobering because the majority of future population growth is expected to occur in small- and medium-sized cities. The bivariate results also suggest a threshold beyond which the largest cities (i.e. megacities) become more efficient in their use of energy and where increases in pollution may be proportionate with increases in population. Such effects might be expected by the post-industrial shift of environmental burdens away from the largest cities to smaller and less developed cities (McGranahan, 2007; Sassen, 2001).
Secondly, the regression results affirm the presence of complex income scaling effects in 2005. That is, the coefficients on GDP per capita indicate that affluence is positively associated with pollution at the low end of the income spectrum, while affluence in the middle range of the income spectrum is negatively associated with pollution. The significant cubic terms for the global sample indicate that the direction of the affluence–pollution relationship reverses at the upper end of the income spectrum, where affluence is again positively associated with pollution. The results are largely consistent with expectations of the UET framework with respect to contemporaneous variation in environmental degradation across cities (McGranahan, 2007).
Nevertheless, an examination of the bivariate relationship between GDP per capita and the predicted values for each city suggests that the middle portion of the curve is not declining so much as it is relatively flat in slope, compared with steeper slopes at the lower and higher ends of the income spectrum. If the cross-sectional findings accurately represent an underlying structural relationship between affluence and urban air pollution, economic growth is not likely to bring much of any reduction in environmental impact for the majority of middle-income cities as posited by ecological modernisation theory, and could be associated with large relative increases in pollution in the least developed and most developed cities.
Thirdly, the results indicate that characteristics of the local economy are extremely important in explaining variation in pollution. The coefficients for
The negative coefficients for
Fourthly, the results hint at a role for urban form in ameliorating urban air pollution. Higher population density is significantly associated with lower aggregate emissions for all models, after controlling for other factors such as economic development and population size. Dense cities may require less travel from their residents and businesses, may require less energy to heat and cool residences, and may be home to less-polluting economic activities than more sprawling cities (even after the shares of emissions from each sector are held constant) (Gonzalez, 2005).
It is more likely that density may be a proxy for the presence of urban slums, in which residents consume less energy than in other forms of urban development (Jorgenson et al., 2010). Indeed, the density coefficients are larger in cities from developing nations than for cities in developed nations, and largest in cities from least developed nations where slums are most likely to be present. Additional analysis is required further to examine these relationships across sectors and over time using more extensive measures of urban form (for example, Schneider et al., 2005) and to test for potentially confounding effects of urban slum development (Jorgenson et al., 2010).
The population growth rates are significantly and negatively related to air pollution in 2005. That is, faster-growing cities are associated with fewer contemporaneous emissions, while slower-growing (or declining) cities are associated with higher contemporaneous emissions, all else equal. These results hold after controlling for affluence and size, indicating the presence of some factor associated with declining or slowly growing cities. Many of the fastest-declining cities among the developed nations are in industrial regions of Russia, Ukraine and Finland. Thus, the population growth variable may proxy for an idiosyncratic effect of industrial development on urban emissions, even after controlling for economic productivity and the share of emissions from industrial activity.
Climate does not explain much variation in emissions across urban areas, although the negative coefficients are statistically significant across the full sample of cities. These findings suggest that tropical cities may have slightly lower emissions, all else equal. The climate variable may pick up the historical influence of colonisation and exploitation in tropical cities, or it may proxy for the presence of concentrated poverty as with the density variable. The climate effect is relatively weak, however, in a model that also includes affluence and density variables.
Limitations
The analysis may suffer from measurement bias and thus the results should be viewed as preliminary and exploratory. Measurement error derives from the use of global spatial datasets, which were constructed with national-level data and ‘downscaled’ to the local level. The EDGAR team allocated some emissions categories by population where information about the location of polluting activity was unknown, such as for rail transport and other non-road vehicles, from residences, from inland navigation and from electricity transmission. Thus, the effect of population on pollution may appear stronger than in reality, especially in less developed cities where urban slum formation may exhibit a confounding effect on energy consumption and associated emissions (Jorgenson et al., 2010).
In addition, the emissions data are ‘direct’ emissions from within urban boundaries and are not adjusted downward for export-related activity that occurs within city bounds or adjusted upward for urban-related activity that occurs outside city bounds. The research and policy literature has debated how best to attribute emissions to cities (Satterthwaite, 2008). Many scholars argue that consumption-based estimates are necessary to capture the full resource ‘footprint’ of cities (Dhakal, 2010; York et al., 2003a). Such approaches require extensive data inputs beyond what could be performed for a global analysis. This approach could be considered a ‘partial footprint’, such as has been calculated for air pollution from megacities (Gurjar et al., 2008) and for carbon emissions from US metropolitan areas and selected large global cities (Brown et al., 2008; Sovacool and Brown, 2010). As others have noted (Butler et al., 2008; Gurjar et al., 2008), further work is needed to map more closely the geography of emissions within individual countries. Recent effort to create fine-scale datasets of observed pollution concentrations based on satellite imagery will be helpful for verifying the distribution of emissions and developing forecasting models (van Donkelaar et al., 2010).
Additional measurement error derives from the use of other modelled datasets that provide the best available indicators of local conditions but are likely to contain non-trivial measurement error. Nevertheless, the regression results are plausible and fit with prior expectations, indicating that the datasets function credibly for high-level comparative analysis.
This cross-sectional analysis is limited in its ability to discern causation. The results are suggestive of causal linkages, but further modelling would be required using a panel dataset to examine the extent to which changes in the explanatory variables are associated with changes in urban air pollution. At the least, the present model allows investigation of correlates of urban air pollution, which provide context for further analysis and policy-making.
Related, the cross-sectional nature of this analysis precludes investigations of dynamic interactions between cities and the changing role of cities within the global economy. Contemporaneous pollution levels reflect the specialisation of economic activities within a global network of cities and the historical displacement of polluting economic activity away from developed cities to less developed cities (McGranahan, 2007; Sassen, 2001).
Finally, the model is underspecified in other ways. For instance, data limitations precluded including factors that are likely to relate to pollution such as a city’s age distribution, governance arrangements or environmental sensibilities.
Discussion
This analysis was designed to shed light on the impact of urbanisation on environmental outcomes. The logic of IPAT suggests that society might achieve its sustainability goals despite population or income growth if consumers could reduce their resource demands and/or producers could become more resource-efficient in their production activities (Ekins, 1997; Weisz and Steinberger, 2010). The results here affirm that city-level pollution is likely to increase with population growth, and at best only decrease slightly with income growth, placing the onus for pollution reduction on consumption and production efficiencies that are likely to come from technology improvements. In addition, the analysis indicates that the share of emissions from the energy sector is a particularly strong correlate of urban pollution. If policy-makers have aspirations for improving living standards globally, then policy-makers must also focus on reducing the emissions intensity of economic activity within cities.
Nevertheless, Jevons’ paradox suggests that reductions in environmental impact are unlikely to result from improvements in technology alone (Alcott, 2005). Improvements in energy technologies reduce the cost of using energy, which can spur increases in total consumption and associated environmental impact, necessitating an alternative or complementary strategy for reducing environmental impact. For instance, the analysis suggests that there may be some pollution reduction benefit from compact urban development. Yet, compact development may increase exposure to urban pollution if it restricts population growth to areas near pollution sources. In addition, powerful business interests aligned in support of sprawling development forms are likely to resist efforts to control development patterns and to improve energy efficiencies within cities (Gonzalez, 2005).
Further analysis could expand this research in several directions. First, researchers and policy-makers are concerned over the influence of coming urbanisation on local air pollution and human health in poor cities. Additional analysis could focus specifically on a subset of very poor cities. Secondly, the analysis could be extended to look at emissions over time, or across different sectors. The EDGAR dataset has gridded data dating in five-year increments to 1970 and for 14 source categories. This analysis serves as a first pass over total emissions in the latest year available, 2005. The extended analysis might employ more sophisticated spatial analysis tools, such as geographically weighted regression. The current analysis should be merged with parallel analyses of greenhouse gas (GHG) emissions from global cities, in order to identify areas where policy might deliver co-ordinated reductions in emissions of GHGs and urban air pollutants. Further research is needed to examine how different forms of urbanisation may relate to urban pollution and human health. Future work could also examine the subset of cities where data are available on their interconnectedness to reach a better understanding of how a city’s role in the global economy and regional hierarchies may influence its pollution levels (Derudder et al., 2003).
Footnotes
Acknowledgements
The author gratefully acknowledges the feedback and suggestions provided by colleagues and two anonymous referees.
Notes
Funding Statement
This research was funded by the Institute for Sustainability Research, Education, and Policy at the George Washington University and completed at the George Washington Institute for Public Policy.