Effects of changing population or density on urban carbon dioxide emissions

Abstract

The question of whether urbanization contributes to increasing carbon dioxide emissions has been mainly investigated via scaling relationships with population or population density. However, these approaches overlook the correlations between population and area, and ignore possible interactions between these quantities. Here, we propose a generalized framework that simultaneously considers the effects of population and area along with possible interactions between these urban metrics. Our results significantly improve the description of emissions and reveal the coupled role between population and density on emissions. These models show that variations in emissions associated with proportionate changes in population or density may not only depend on the magnitude of these changes but also on the initial values of these quantities. For US areas, the larger the city, the higher is the impact of changing its population or density on its emissions; but population changes always have a greater effect on emissions than population density.

Fig. 1

Conventional approaches for investigating urban emissions. a Urban scaling: the scaling relationship between CO₂ emissions (C) and population size (P). The dashed line represents a power-law fit (Eq. (1)) with an exponent β = 0.48 ± 0.01. We observe that this model underestimates the emissions for large population sizes. b Per capita density scaling: the scaling law between CO₂ emissions per capita (C/P) and population density (P/A). The dashed line is a power-law fit (Eq. (2)) with an exponent α = −0.79 ± 0.01. In both plots, each dot is associated with a US urban unit obtained from the city clustering algorithm (see Methods) and all quantities are expressed in base-10 logarithmic scale. Emissions are measured in tonnes of CO₂, population in raw counts, and area in square kilometers

Fig. 2

The interplay between population and area on CO₂ emissions. a Scatter plot of the observed values of CO₂ emissions (C) and those predicted (C_P) by the Cobb–Douglas model (Eq. (3) with β_P = 0.31 ± 0.01 and β_A  = 0.45 ± 0.03). This model is a significantly better fit when compared with the urban scaling and the per capita density scaling models (Supplementary Fig. 2). b A contour plot of Eq. (3) as a function of P and A on logarithmic scale. The straight isolines/isoquants show how population and area must change in order to keep the emissions unchanged. c Scatter plot between the observed and predicted CO₂ emissions obtained from the translog model (Eq. (5) with β_P = 0.28 ± 0.02, β_A = 0.14 ± 0.05, and β_C = 0.07 ± 0.01). This model further refines the goodness of the predictions (Supplementary Fig. 2), particularly reducing the bias in urban areas with high emissions. d A contour plot of Eq. (5) as a function of P and A. We note that the isolines/isoquants of this model are not straight lines as those from the Cobb–Douglas model. We have employed base-10 logarithmic quantities in all panels