Urban emission scaling — Research insights and a way forward

Much of human progress today can be attributed to the conglomeration of skilled people to a certain space at a given time. Modern day cities stand tall as epitomes of human innovative power and co-creation capacity. However, similar to the “Yin and Yang” depicted in the ancient Chinese philosophy, cities as dynamic systems do exhibit their “Yin” side. Urbanization, increasing prosperity, and (physical, social) infrastructure systems lead to an unprecedented increase in global energy and resource consumption. Much of anthropogenic transformation of Earth's environment in terms of environmental pollution at local level to planetary scale in the form of climate change is currently taking place in cities. Therefore, the ultimate fate of humanity predominantly lies in the hands of technological innovation, urbanites' attitudes towards energy and resource consumption and development pathways undertaken by current and future cities.

The complex nature of cities lead to the quest for identifying appropriate theories and models which capture urban dynamics more appropriately. Drawing inspiration from the advances in the complexity sciences, a new science of cities evolved in the recent years which acknowledged cities as a result of various bottom-up evolutionary processes (Batty, 2012). Amongst others, urban scaling (Bettencourt et al., 2007) has been identified as a crucial part in understanding urban dynamics. Urban scaling depicts that certain urban indicators (I) such as GDP, crime rates, infrastructure, and innovation can be described through a power–law relationship as a function of urban population (P) as follows

P ∼ I β .

The exponent (β) determines the efficiency of large cities in comparison to smaller cities with respect to the urban indicator in question. While super-linear scaling (β > 1) depict increasing economies of scale, sub-linear scaling (β < 1) depicts decreasing economies of scale. Initially limited to indicators concerning socio-economic and infrastructure aspects of cities, studies on urban scaling have been later extended to environmental indicators such as greenhouse gas (GHG) emissions. Such studies are crucial since they might enable researchers and local policy makers to understand the influence of on-going urbanization on GHG emissions and identify the intricacies of urban emission patterns.

However, the urban scaling approach still has some issues to be resolved. Recent studies aiming at understanding the scaling of urban emissions (mostly limited to cities in developed regions) have depicted contradictory results (Gudipudi et al., 2019). In a previous commentary, Louf and Barthelemy (2014) showed that such results might be the consequence of inappropriate estimation of the quantity of urban transport emissions and the lack of proper definition of city itself as showed by Arcaute et al. (2014). Here, we go further and state that there are other (equally) important factors that determine the scaling of urban emissions. This commentary aims at highlighting these issues, their policy implications, and the way forward for urban emission scaling studies.

The foremost issue is the emission accounting inventory used to quantify urban emissions. More precisely, the scope of emission inventory used in the analysis—i.e. scope 1, 2, or 3. While emissions reported under scope 1 consider GHGs originated within a given urban boundary, scope 2 includes emissions from the electricity and heating energy imported from energy generating units which are beyond the urban boundary and scope 3 includes embedded net emissions in the supply chain of goods and materials (imported and exported elsewhere)—commonly referred as consumption-based emission accounting (for further details see https://ghgprotocol.org/sites/default/files/standards/GHGP_GPC_0.pdf). While there is a general consensus that consumption-based emission accounting is desired for analyzing emissions, most of the urban emissions scaling studies are limited to scope 2 or scope 1 inventory. This is largely due to a lack of emission data available in the embedded life cycle emissions. Urban emission scaling studies so far depicted that large cities in developed regions are either not very efficient with a slope (β) ≈ 1 (Fragkias et al., 2013) or relatively more efficient with a slope (β) <1 (Rybski et al., 2017). However, from both a scientific and policy perspective, it is crucial to analyze the emission efficiency of large cities in these regions using scope 3 emissions since most of these predominantly service sector cities import their goods and materials which are often produced in cities in developing regions (Satterthwaite, 2009) where local policy makers have limited control over minimizing such embedded emissions.

The second issue that needs to be addressed is the level of spatial detail in which emission inventories are often reported. Some emission inventories such as the EDGAR database (for more details: http://edgar.jrc.ec.europa.eu/) use certain proxies (such as population and road density) to downscale national emissions to spatial scales (0.1 degree * 0.1 degree), while others such as the Vulcan database (for more details: http://vulcan.project.asu.edu/) report bottom-up emissions based on the energy consumed within various sectors at county level in the USA. Although, the gridded emission datasets enable researchers to observe urban scaling under different city definitions (as shown in Cottineau et al., 2017), the results can be influenced by the proxies used while downscaling national emissions. On the contrary, unlike the Vulcan database, most of the bottom-up emission inventories are limited to individual cities which usually are confined to the municipal boundaries and therefore fail to give a complete picture while using different city definitions. Addressing this issue is of particular interest since using the same Vulcan emission inventory, Fragkias et al. (2013) showed that large cities are (almost) as efficient as smaller cities while Oliveira et al. (2015) depicted that large cities are less efficient. In line with our arguments in the previous paragraph, a complete picture with regard to emission efficiency of large cities is possible only through using bottom-up scope 3 emission inventories translated to gridded scale.

The final issue which we want to bring to the limelight is often disregarded in urban scaling studies, i.e. the regression method used (Leitão et al., 2016). In Figure 1, we illustrate how the slope (β) changes with the application of five different regression methods on the data of one of the first studies published on urban emission scaling by Fragkias et al. (2013) for 933 core-based statistical areas (CBSAs) in the USA. Following the commonly used ordinary least squared (OLS) regression in urban scaling studies one can interpret from Figure 1 that large cities are (almost) as emission efficient as smaller cities while all the other four regression methods say the opposite (i.e. large cities are inefficient in comparison to smaller cities). Moreover, OLS is not appropriate while understanding complex interactions amongst various dependent and independent variables as shown in Figure 2. Let us assume we are interested in the scaling of emissions as a function of population size and understand the intrinsic factors determining the emission efficiency of large cities. One approach to achieve this is to take a detour by identifying a chain of scaling relations between GDP and population, energy and GDP, and finally emissions and energy. Such an exercise will determine which of the three efficiencies, i.e. affluence (exponent β), energy intensity (exponent α), and carbon intensity (exponent γ), will have a stronger influence in determining emission efficiency of large cities compared to smaller ones further details in Figure 2 below. In Gudipudi et al. (2019), an “Urban Kaya Scaling” has been derived which describes these complex interactions. However, for the “Urban Kaya Scaling” to be formally valid, one has to use reduced major axis (RMA) regression method since the correlation coefficients among these interactions are not the same while swapping the dependent and independent variables when using OLS. How should one proceed when the choice of regression method determines the efficiency of large cities? This finding is not new in empirical studies since the results (in many cases) vary depending on the choice of method. However, the practical implications while transferring urban scaling science to policy making is of profound significance.

OPEN IN VIEWER

Figure 1.

Are large cities more emission efficient in comparison to smaller cities? Scaling of total emissions with the population size for 933 core-based statistical areas in the USA. Data are obtained from Fragkias et al. (2013). While applying the commonly used regression method in urban scaling studies, i.e., ordinary least squared regression (OLS), we obtained a slope (β) = 0.93 which depicts that large cities are (almost) as efficient as smaller cities. The slopes derived from other regression methods such as inverse OLS, orthogonal regression (OR), OLS mean, and reduced major axis (RMA) are 1.36, 1.15, 1.14, and 1.13, respectively, depicting that large cities are not emission efficient in comparison to smaller cities. For further details methodological approaches of these regression methods please refer to (Jogesh Babu and Feigelson, 1992). The gray dotted line has a slope of 1 and is shown to illustrate regression lines above and below.

OPEN IN VIEWER

Figure 2.

Complex interactions among scaling relationships identified in Gudipudi et al. (2019) where “C” is CO₂ emissions, “P” is population, “G” is GDP, and “E” is energy consumed at city level. The factors leading to emission efficiency of large cities (i.e. slope φ ) as derived in the “Urban Kaya Scaling” is given by the product of the other slopes (i.e. φ =βαγ). Further details are present in Gudipudi et al. (2019).

Urban scaling studies provide insights in determining what is general and specific to a given city. However, translating these scaling studies to urban (and regional) policy is not straightforward especially with regard to emission scaling studies until the aforementioned issues are addressed adequately. Moreover, majority of the urban scaling studies usually focus on many urban areas at a single time step. Urban policy making based on such results and their consequences for future urban development still remains an open question. For instance, let us assume a theoretical possibility that the energy supply grid of an urban system relies solely on renewable and sustainable resources. Now, the scaling hypothesis (i.e. emissions as a power law of population size) will not be valid since the scope 1 emission from the energy consumed from renewable resources will be zero. However, there will be certain emissions embedded in the lifecycle of these renewables, which usually are beyond the scope of the urban policy makers. Therefore, from a policy perspective, resolving these issues is essential especially for cities in the developing world as these regions are projected to have higher urbanization rates coupled with increasing affluence while lacking adequate technologies/infrastructure—a lethal combination of the “Yin” side of urbanization that leads to increased resource consumption and GHG emissions.