Multi-Scale Land Use Impacts on Fossil Fuel-Related CO$_2$ Emissions in the United States

TL;DR

The paper addresses spatial heterogeneity in fossil-fuel CO2 emissions in the United States by leveraging the Vulcan FFCO2 dataset to study land-use effects at multiple scales (CBG neighbourhoods and CBSA metros). It adopts a causal inference framework based on a generalized propensity score for continuous treatments and a doubly robust estimator, with BART for confounder modelling and extensive propensity-score balancing. The authors find that aggregate FFCO2 scales with population size at the metropolitan level (slope ≈ $0.92$, $R^2 ≈ 0.69$), but neighbourhood-scale scaling is largely absent; transportation emissions show allocation-related biases, while scope 2 electricity emissions exhibit sublinear scaling. The study advances policy-relevant understanding by linking detailed land-use features to sector-specific emissions and highlights limitations of current allocation methods, offering a robust, data-driven approach for informing urban climate planning and multi-scale policy design. These findings underscore the value and challenges of causal, high-resolution urban emission analyses for targeted interventions.

Abstract

Anthropogenic greenhouse gas (GHG) emissions exhibit spatial variation owing to differences in development patterns, local climate, economic composition, energy sources, and other factors. Many of these factors - and therefore their contribution to GHG production - are influenceable through spatial planning and economic policy. Recent advances in environmental data reporting and climate flux measurement have produced high fidelity GHG emissions estimates at detailed spatial and temporal resolutions. Using one such dataset (Vulcan v3.0 1-km gridcell estimates of fossil-fuel CO2 for the U.S.), we explore the relationship between land use features and CO2 emissions. Analysis is conducted at multiple scales (neighbourhoods and metropolitan areas) to explore scale law effects. Using a data-driven propensity score approach, we develop doubly robust causal estimands for the effects of multiple land use features on CO2 emissions by sector. Preliminary results suggest that per capita transportation emissions are not significantly affected by local population density after controlling for metropolitan population density factors. However, results are likely influenced by the way transportation emissions are allocated in the Vulcan dataset. Results for scope 2 residential electricity and non-electricity energy are also considered in the study.

Figures (10)

• Figure 1: Annual gasoline consumption with respect to urban density Newman_Kenworthy_1989.
• Figure 2: Grid cell-CBG spatial imputation
• Figure 3: Causal graph for variable selection
• Figure 4: Metropolitan scaling effect by sector
• Figure 5: Neighbourhood scaling effect by sector