Growing global temperatures driven by greenhouse gas (GHG) emissions, primarily from fossil fuel combustion, necessitate understanding the factors driving energy demand and emissions. Existing cross-country analyses often focus on the STIRPAT framework, emphasizing population, affluence (typically measured by GDP), and technology. While GDP's role as a major determinant is acknowledged, the debate on decoupling GDP from emissions remains inconclusive. This study addresses the need for exploring additional factors, particularly the often-overlooked influence of built structures (settlement and transport infrastructure patterns) on national-level energy demand and CO₂ emissions. Previous urban-scale research highlights the impact of built structures on resource demand, but a lack of consistent national-level indicators has hindered similar analyses at this scale. The paper aims to fill this gap by developing and applying national-level indicators to assess the influence of built structures alongside conventional factors in predicting energy demand and CO₂ emissions across 113 countries.

The literature review highlights the varied findings regarding the relationship between population density and CO₂ emissions. Some studies suggest an inverse relationship, while others find no effect. Studies at the urban scale extensively demonstrate the influence of population density and spatial layout on cities' resource demand. However, national-level analyses often neglect material stock patterns (spatial patterns of settlements and infrastructures), despite the acknowledged link between the accumulation of material stocks (buildings and infrastructure), energy demand for heating, cooling, lighting and mobility, and associated GHG emissions. A systematic review of the literature revealed a paucity of studies considering material stock patterns at the national level, primarily due to the lack of suitable, comparable indicators.

This study develops three types of national-level indicators to characterize material stock patterns: (1) indicators representing the area of built-up land (as a fraction of inhabited land and per capita), spatial clustering, form, and distribution; (2) road density indicators for urban and rural areas and the ratios between them; and (3) similar railway indicators. These indicators were quantified for 113 countries, encompassing 91.2% of the world's population and 97.3% of global GDP. The analysis uses two dependent variables: yearly total per capita final energy consumption (TFC) and yearly per capita CO₂ emissions (CO₂). These variables were tested against conventional factors (GDP per capita, population density, urban population share, heating-degree days, and gasoline price) using bivariate and semi-partial correlations. A multivariate lasso analysis was employed to select variables for multivariate models predicting cross-country patterns of TFC and CO₂. This approach addressed potential overfitting and multicollinearity. Cross-validation was used to determine the optimal model by minimizing out-of-sample prediction error. Data sources included UN Statistics Division, World Bank, IEA, and Global Carbon Project databases, using data averaged over 2015-2020 to reduce random fluctuations. Spatial indicators were derived from Copernicus Global Land Cover Service data, OpenStreetMap data, and EUROSTAT GISCO archive. The study also defines a proxy layer for inhabited land, which serves as a more suitable area reference than total national territory, particularly for countries with varying levels of inhabitable land.

Several key findings emerged from the analysis. Bivariate correlations revealed that several material stock pattern indicators exhibited similar correlations with TFC and CO₂ as conventional factors. The area of built-up land per capita (BLcap) demonstrated a strong positive correlation with both TFC and CO₂, second only to GDP in predictive power. Semi-partial correlations, controlling for GDP and population density, further confirmed the additional explanatory power of many material stock pattern indicators. Multivariate lasso analysis identified BLcap as the second-most important predictor for both CO₂ and TFC after GDP. The lasso analysis also showed that various material stock pattern indicators were selected much earlier than conventional factors like population density. Models incorporating material stock pattern indicators demonstrated superior predictive power compared to models using only conventional factors for both CO₂ and TFC, as measured by in-sample and out-of-sample goodness of fit and BIC. Even models using only material stock pattern indicators achieved reasonably good predictive performance. Population density played a smaller role than expected, whereas many aspects of material stock patterns significantly impacted cross-country differences in energy demand and CO₂ emissions.

The findings demonstrate the significant role of built structures' extent and spatial patterns in determining national-level energy demand and CO₂ emissions. These results support the generalizability of urban-scale findings to the national level. The developed national-level indicators effectively capture crucial characteristics of material stock patterns that influence energy use and emissions. The strong predictive power of BLcap underscores the importance of the area of built-up land in energy demand and CO₂ emissions, primarily due to energy use during construction and operation of buildings and infrastructure, as well as increased transport energy demand resulting from larger floor sizes and longer distances. The findings suggest that limiting built-up area per capita could be a valuable policy strategy for mitigating GHG emissions. The study acknowledges collinearities between material stock pattern indicators, highlighting the need for future research employing refined study designs to address temporal and spatial dynamics and elucidate causal pathways.

This study demonstrates that the extent and spatial patterns of built structures are crucial in determining national-level energy demand and CO₂ emissions. The area of built-up land per capita is a particularly powerful predictor. These findings highlight the importance of incorporating material stock patterns into national-level analyses of energy use and emissions. Future research should focus on more in-depth analysis of the temporal dynamics of these relationships and explore the diverse causal mechanisms involved. Furthermore, leveraging high-resolution maps of changes in built structures over time offers significant potential for improving model accuracy and informing climate change mitigation strategies.

The study acknowledges limitations stemming from the aggregation of spatially explicit data to the national level, which inevitably leads to information loss. Data availability also posed limitations; for instance, the reliance on averages over several years to reduce annual fluctuations might mask some short-term dynamics. Future work could address these limitations by exploring sub-national level analysis and incorporating more detailed temporal data.