Written by Helmut Haberl
Which factors affect national economies' resource demand and emissions? The question is of key importance, given that global heating is progressing rapidly while progress in climate-change mitigation policies fall dramatically short of the efforts required to reach agreed targets (e.g. the 1.5°C 'Paris' goal). Many cross-country analyses of that question conducted so far used the venerable IPAT (Impact = Population x Affluence x Technology) approach to study this question, which was more recently further developed into the STIRPAT (Stochastic Impacts by Regression on Population, Affluence and Technology) framework (Dietz et al. 2020). The national level is particularly relevant because this is where many decision-making processes occur and policies are formulated and executed.
A main determinant of resource use and emissions considered in STIRPAT as 'Affluence' is Gross Domestic Product (GDP), i.e. the value of goods and services generated by a national economy within one year. A large debate rages whether GDP can be 'decoupled' from resource use and emissions, e.g. through more efficient technologies. 'Relative decoupling' means that GDP grows at higher rates than resource use and emissions, which has been found in many instances. But relative decoupling is grossly insufficient for ambitious climate targets, which require yearly absolute reductions of emissions of 5-10% per year over the next decades to reach net-zero globally around mid-century (IPCC, 2022). Some countries have in recent years implemented climate-change mitigation policies, which has in some cases resulted in absolute reductions of their CO2 emissions (Quéré et al., 2019). 'Absolute decoupling' denotes a situation where resource use and emissions decline in absolute terms while GDP continues to grow. In a systematic review of >800 empirical studies of economy-wide decoupling (Haberl et al., 2020), we had found few cases of absolute decoupling, and no examples where reduction rates consistent with ambitious climate targets had been achieved.
Other potential determinants of energy demand and emissions so far considered in STIRPAT studies include population density, fuel prices, the urbanization rate, heating-degree days (a proxy of heating-energy demand) and similar indicators. Those studies usually found the influence of these factors to be a lot weaker and less consistent than that of GDP.
Extent and patterns of built structures are notably absent from STIRPAT-style studies of economy-wide determinants of resource use. This may seem surprising, given that the influence of 'urban form' (i.e. density and spatial arrangements of settlements and urban infrastructures) are a well-known determinant of cities' resource requirements, as also acknowledged in IPCC reports (Seto et al., 2014). This should not be surprising: Accumulation of materials in built structures requires massive amounts of emission-intensive products such as steel-reinforced concrete, mortar, bricks, timber, plastics, gravel, sand and so on. Moreover, use of built structures again requires huge amounts of energy, e.g. for lighting, heating or vehicles (Fig. 1).
Workflow of the study
Indicators representing built structures could so far not be considered in STIRPAT-style analyses due to a scale mismatch. While maps of built structures reveal fine-grained spatial detail and resolutions are continuously on the rise (meanwhile down to 10x10 pixels), such rich detail cannot be included in cross-country analyses. These studies require that the massive information content of building and infrastructure maps is aggregated to the national level. The challenge is to do this in a manner that preserves key information on extent and patterns of built structures. STIRPAT-style analyses can deal only with a limited number of such indicators, which makes this task even more challenging.
Fig. 2 summarizes the workflow of our study, which aims to remedy that situation. More detail is available in the paper (https://doi.org/10.1038/s41467-023-39728-3). After a lot of trial & error, as well as tedious GIS work, we arrived at a set of six indicators of the extent and spatial pattern of built-up land, and five indicators each for the distribution of roads and railways. These include simple indicators describing the extent (e.g. built-up land per capita) and density (e.g. fraction of a nation's inhabited land area covered with buildings, road or railway density) of built structures. We also included more sophisticated measures of the dispersion and compactness of built-up land, urban density, or urban and rural railway and road densities. The full indicator set is described in Table 1 in the paper (https://doi.org/10.1038/s41467-023-39728-3); the online Supplement (SI file) contains detailed descriptions and maps describing the indicators and patterns. Data (Data file) and code (Code file) are also freely available as zenodo archives.
The design of the statistical analyses presented us with another set of difficult challenges, most importantly resulting (1) from the large number of potential determinants, and (2) their high level of collinearity. We therefore experimented with a large number of possible analysis methods, eventually deciding for the following approach:
In a first step, we embarked on simple bivariate correlation analyses (see upper part of Fig. 2 in the paper; https://www.nature.com/articles/s41467-023-39728-3). Bivariate analyses show how strongly each potential determinant is correlated with each country's per-capita total final energy use (TFC) and CO2 emissions (CO2). Here we mostly found the expected patterns; still, some results were surprising, for example the very strong correlation between built-up land per capita (BLcap) and both TFC and CO2, as well as the very small role of populations density (which had been identified as an important parameter in some - but not all - previous studies) and fuel prices.
In a next step, we conducted semipartial correlations controlled for GDP and population density. This means that the part of each built-structure indicator correlated with GDP and population density is removed, and only the strength of the linear correlation between energy or CO2 with the remaining part of that indicator is analyzed. This unveiled some underlying patterns invisible in the bivariate analyses, e.g. that fuel prices or urban road density do play a role in such a setting, even though they were not found to be strongly correlated in the bivariate setting. These results are shown in the lower part of Fig. 2 in the paper; https://www.nature.com/articles/s41467-023-39728-3).
In a final step, we employed an advanced variable selection approach termed 'lasso' (least absolute shrinkage and selection operator). This multivariate method is used for automatically performing variable selection in linear regression models. Lasso overcomes drawbacks with overfitting and is particularly useful in situations where optimal models are to be developed from large numbers of multi-collinear potential determinants. In this procedure, a parameter is introduced to penalize model complexity. The algorithm starts at a high penalty, which is gradually reduced to include more and more variables, until an optimal statistical model is identified. In a lasso analysis, the order in which variables are selected helps identifying the factors that most strongly affect the cross-country pattern, in decreasing order. Results of the cross-validation lasso analysis are shown in Table 2 in the paper (https://www.nature.com/articles/s41467-023-39728-3), alternative approaches (which yielded largely similar results) are presented in the online Supplement (SI file).
Insights and conclusions
The first major insight of our study is that built structures indeed play an important role in co-determining cross-national patterns of per-capita energy use and CO2 emissions. Our indicators of extent and spatial patterns of built structures have a strong, additional explanatory and predictive power, beyond GDP and other indicators considered so far in STIRPAT-style studies.
Despite the unavailable loss of information resulting from aggregating to the national scale, our indicators are able to represent important characteristics of built structures at the national level. This means that they can (and hopefully will) be used to study the importance of built structures for other aspects of national economies' resource use, emissions or waste flows than those analyzed in our paper. This will help develop much stronger models and predictions of national-level energy use and emissions, and enrich the debate on a possible decoupling of GDP from resource use and emissions.
Second, insights from urban-level studies do indeed translate to the national scale. Empirical urban studies have consistently shown that urban form and transport infrastructures affect travel demand, and thereby GHG emissions (Ewing & Cervero, 2010). These effects are obviously sufficiently strong to show up at the national level as well.
Third, the indicator with the strongest and most consistent predictive power across all analyses is the area of built-up land per capita (BLcap). BLcap emerges as the second-most important determinant in most analyses, including those considering the GDP effect. This is plausible because infrastructures and buildings require energy (1) for being built as well as (2) during their use-phase (e.g. energy use for heating or vehicles, as well as resources required for maintenance).
Higher BLcap also means larger floor size and longer distances between locations, all of which raises energy demand and results in higher CO2 emissions, given currently prevailing fossil-fuel dominated energy systems. Moreover, the higher energy demand is, the larger the difficulties (e.g. in terms of costs or material resource requirements) will be to decarbonize the energy system.
Our results have implications for achieving ambitious climate targets: (1) limit new land-demand of built structures, (2) follow recommendations from urban studies, as they will translate to the national level. Overall, area and patterns of built structures emerge as an important, so far underappreciated entry point for reducing energy demand and CO2 emissions.
Dietz, T., Shwom, R.L., Whitley, C.T. (2020). Climate change and society. Annu. Rev. Sociol. 46, 135–158, https://doi.org/10.1146/annurev-soc-121919-054614
Ewing R. & Cervero, R. 2010. Travel and the built environment. J. Am. Plan. Assoc. 76, 265-294, https://doi.org/10.1080/01944361003766766
Haberl, H. et al. 2020. A systematic review of the evidence on decoupling of GDP, resource use and GHG emissions, part II: synthesizing the insights. Environ. Res. Lett. 15, 065003, https://doi.org/10.1088/1748-9326/ab842a
IPCC, 2022. Climate Change 2022 - Mitigation of Climate Change. Intergovernmental Panel on Climate Change, Geneva
Quéré, C.L. et al., 2019. Drivers of declining CO2 emissions in 18 developed economies. Nat. Clim. Chang. 9, 213–217, https://www.nature.com/articles/s41558-019-0419-7
Seto, K. C. et al., 2014. Human settlements, infrastructure and spatial planning. In Climate Change 2014: Mitigation of Climate Change. In: Edenhofer, O. et al. (eds.). Working Group III contribution to the IPCC Fifth Assessment Report (AR5) of the Intergovernmental Panel for Climate Change, Cambridge Univ. Press, pp. 923–1000