Built structures influence patterns of energy demand and CO₂ emissions across countries

The study investigates to what extent national-scale built structures—settlement patterns and transport infrastructures—co-determine per-capita energy use (total final energy consumption, TFC) and territorial CO2 emissions across countries. While cross-country analyses typically emphasize population, affluence (GDP), and technology as in IPAT/STIRPAT frameworks, the role of material stock patterns has been underexplored at national scale due to data limitations. Urban research suggests built form and infrastructure strongly influence energy demand and emissions, raising the question whether similar relationships hold at national level and whether indicators of built structures add explanatory power beyond conventional factors such as GDP, population density, urbanization rate, climate (heating-degree days), and energy prices (gasoline price). The purpose is to develop national indicators of built structures, quantify them for 113 countries, and test their ability to explain cross-country variation in TFC and CO2 alongside conventional determinants.

Prior literature emphasizes GDP as a major driver of energy use and GHG emissions within IPAT/STIRPAT analyses, with ongoing debate over the possibility of absolute decoupling. Evidence of recent emissions reductions in some developed economies exists, but is insufficient to meet ambitious climate targets. Population density shows mixed associations with energy use and emissions across studies, with some finding inverse relationships and others no effect. Urban-scale research consistently links urban form, density, and infrastructure to energy demand in buildings and transport. However, national-level analyses rarely include material stock patterns; a systematic review found only one national study considering spatial distribution of cities for transport emissions. This gap reflects a scale mismatch between fine-grained spatial datasets and national comparative analyses, motivating the development of consistent, aggregatable indicators of built structures for cross-country studies.

Design: Cross-sectional analysis for 113 countries (covering 91.2% of world population and 97.3% of global GDP). Dependent variables: per-capita total final energy consumption (TFC) and per-capita territorial CO2 emissions from fossil fuel combustion and cement production (CO2). Conventional factors: GDP per capita (constant 2015 USD), population density (DENS), urban population share (UPOP), heating-degree days (HDD), and pump price for gasoline (PGAS). All extensive variables are expressed per capita. Where possible, values are averaged for 2015–2020; TFC is from IEA balances (latest year 2017). Data record download date: 25 March 2021. Spatial indicators: Derived national-level indicators representing material stock patterns in three groups: (1) built-up land extent and pattern (fraction of built-up land, built-up land per capita, dispersion, monocentricity, compactness, urban population density); (2) road network (total, urban, rural densities and urban-to-rural ratios of lengths and densities); (3) railway network (analogous to roads). Built-up land and urban agglomerations were mapped from the Copernicus Global Land Cover LC100 100 m grid (epoch 2015) after vectorization and a growing-neighborhood clustering approach to delineate agglomerations. Road and railway vectors were sourced from OpenStreetMap via Geofabrik, with minor classes excluded for global consistency; urban/rural distinctions were made by spatial intersection with built-up land features. A custom inhabited land (IH) reference layer was constructed (masking uninhabitable areas by elevation and land cover, buffering built-up features, and including croplands) to normalize area-based indicators. Statistical analysis: Variables were log-transformed for correlation analyses. Bivariate Pearson correlations were computed between each indicator and TFC/CO2. Semi-partial correlations were calculated controlling for GDP and DENS to isolate the additional explanatory component of material stock indicators. Multivariate modeling used lasso regression with cross-validation (K-fold) to select parsimonious sets of predictors and estimate in-sample R² and out-of-sample cross-validated R² (osR²). Benchmark models using only conventional factors were compared with models including both conventional and material stock indicators, and with models including only material stock indicators. Model selection metrics included cross-validated mean-square prediction error (MSPE) and Bayesian Information Criterion (BIC).

• Many material stock pattern indicators are as strongly correlated with TFC and CO2 as conventional factors. In bivariate correlations (log scale):
  • TFC: GDP 0.86; BLcap 0.72; UPOP 0.67; RWDtotal 0.67; RWDrural 0.65; RWDurban 0.59; HDD 0.53; RDrural 0.47; RDtotal 0.46; BLfract 0.44. Negative correlations: BLcomp −0.44; UPdens −0.45; RDurb-rur −0.52; BLmono −0.31.
  • CO2: GDP 0.82; BLcap 0.72; UPOP 0.69; RWDtotal 0.68; RWDrural 0.67; RWDurban 0.56; HDD 0.55; BLfract 0.50; RDtotal 0.49. Negative correlations: RDurb-rur −0.50; UPdens −0.43; BLcomp −0.36; BLmono −0.32.
• Semi-partial correlations (controlling for GDP and DENS) show additional explanatory power from material stock indicators: built-up land per capita (BLcap) and fraction (BLfract) remain positively associated with TFC and CO2; rail and road densities—especially rural—add explanatory power; UPOP becomes inversely associated with TFC and CO2 after controls; PGAS and some road indicators gain importance after controlling.
• Lasso variable selection: GDP is always selected first; BLcap is selected second for both TFC and CO2 and remains in the optimal models. Several material stock indicators enter earlier than population density and persist in optimal models.
• Predictive performance: Models including material stock indicators outperform those with only conventional factors.
  • For TFC, Model A (with material stock indicators) vs Model B (conventional only): R² 0.900 vs 0.851; osR² 0.865 vs 0.833; BIC 84.17 vs 100.32 (lower is better).
  • Models using only material stock indicators (Model C) still achieve strong predictions: osR² 0.65 for TFC and 0.62 for CO2.
• Consistent central result: Built-up land per capita (BLcap) is the strongest predictor after GDP across analyses, indicating that the extent of material stocks per person is tightly linked to national per-capita energy demand and CO2 emissions.
• Population density plays a smaller role than often assumed; urban population density is inversely related to energy and emissions, while higher urbanization shares are positively related in bivariate analysis but reverse sign after controlling for GDP and density in semi-partial analyses for some outcomes.

Findings confirm that national-scale built structures—captured by indicators of built-up land extent and the density/configuration of road and rail networks—are major co-determinants of per-capita energy demand and CO2 emissions, comparable in importance to GDP. The robustness of built-up land per capita (BLcap) across correlation and multivariate analyses suggests a generalizable mechanism: larger material stocks (more floor area and more dispersed settlements with longer travel distances) increase energy use for buildings and transport in fossil fuel–dominated systems, thereby raising CO2 emissions. Translating insights from urban studies to national scale, the developed indicators retain sufficient information despite aggregation to provide additional explanatory and predictive power beyond conventional STIRPAT factors. The results imply that national-level decarbonization strategies should consider the extent and spatial patterns of built structures as policy levers alongside economic, technological, and pricing instruments.

This study develops and quantifies nationally comparable indicators of built structures for 113 countries and demonstrates that material stock patterns significantly explain cross-country variation in per-capita energy use and CO2 emissions, adding substantial predictive power beyond GDP and other conventional factors. Built-up land per capita consistently emerges as the most important predictor after GDP. The indicators enable improved modeling of the GDP–emissions/energy nexus and open avenues to incorporate built structure patterns into national mitigation scenario analyses. Policy implications include considering limits to built-up area per capita and promoting compact, connected development to reduce energy demand and emissions. Future research should address temporal dynamics, causal mechanisms, and interactions among material stock indicators using longitudinal datasets as high-resolution time-series of built structures become available, and refine methods to disentangle collinearities among spatial indicators.

• Aggregation of high-resolution spatial data to national indicators entails an unavoidable loss of information about local patterns and heterogeneity.
• Strong collinearity among material stock indicators complicates disentangling causal pathways in multivariate analyses.
• Cross-sectional design limits inference on temporal dynamics and causality; temporal changes were not the primary focus.
• Variability in OpenStreetMap data completeness across countries necessitated excluding minor classes to improve consistency, which may omit some infrastructure detail.
• The inhabited land (IH) reference layer is tailored for consistency with LC100 and used as an area reference; its spatial accuracy for other purposes was not validated here.
• Energy and emissions data availability constraints (e.g., TFC latest year 2017) required using latest available years or short-period averages, which may introduce timing mismatches.