Multiscale Carbon Burden of Infrastructure in the United States

Technical Summary: Multiscale Carbon Burden of Infrastructure in the United States

Problem Statement
Anthropogenic greenhouse gas (GHG) emissions are spatially heterogeneous, varying with development patterns, climate, and economic structures. Historically, policy-relevant analysis has been constrained by data granularity, often relying on proxy measures like total vehicle kilometres travelled (VKT) and mode-specific emission factors. These approaches obscure within-region heterogeneity and fail to distinguish between consumption-based emissions (attributed to households) and production-based emissions (attributed to the point of generation). Furthermore, existing literature often isolates either macro-scale metropolitan effects or micro-scale neighbourhood design, rarely integrating both to understand scale-dependent relationships. A critical gap remains in understanding the "carbon burden of infrastructure"—specifically, how land use and infrastructure features shape emissions when allocated to their physical point of generation rather than household origin.

Methodology
This study utilizes high-fidelity fossil-fuel CO2 (FFCO2) data from the Vulcan v4.0 dataset, which provides 1-km grid resolution emissions for the United States. The analysis aggregates these data to Census Block Groups (CBGs) and Core-Based Statistical Areas (CBSAs) to align with land use and urban form data from the EPA Smart Location Database.

The study employs a dual-methodological framework to address causal identification and spatial dependence:

1. Doubly Robust (DR) Causal Inference: The author applies Doubly Robust Generalised Propensity Score (GPS) estimation using Bayesian Additive Regression Trees (BART). This approach estimates the Average Treatment Effect (ATE) of continuous land use variables (the "5Ds": population density, land use diversity, roadway design, distance to transit, and destination accessibility) on emissions. The DR estimator balances the probability of treatment assignment and the outcome model, ensuring unbiased estimates even if one model is misspecified. Effects are measured as the change in emissions between the 25th and 75th percentiles of treatment variables.
2. Multi-Treatment Spatial Regression: To capture spatial interactions and scale effects, the study estimates log-linear joint models incorporating both local (CBG) and regional (CBSA) land use variables. The preferred specification is a Heteroscedasticity-robust Spatial Lag of X and Spatial Error Model (SLX+SEM) with a 10-km distance band. This model accounts for residual spatial autocorrelation and treats transportation emissions as production-based, capturing the infrastructure carbon burden hosted by specific communities.

Key Contributions
The paper makes four primary methodological and empirical contributions:

1. Causal Identification: It employs doubly robust estimation to address causal identification in observational data, accounting for both treatment assignment and outcome models simultaneously.
2. Multi-Scale Integration: It integrates land use variables defined at both local (neighbourhood) and metropolitan scales within a consistent model framework to examine how effects operate across spatial hierarchies.
3. Production-Based Allocation: It provides a production-based allocation of transportation emissions, shifting the analytical focus from household travel demand to the carbon burden infrastructure places on communities. This reveals spatial mismatches where urban centres "host" emissions generated by regional mobility.
4. Sectoral Disaggregation: It distinguishes between transportation, residential electricity, and residential non-electricity energy sources to identify differential effects of land use across emission sectors.

Key Results

• Transportation Emissions (Production-Based):
  • Local Density: In the preferred joint spatial model, higher local population density is a strong negative predictor of transportation FFCO2 (approx. -19.5% per 1 SD increase).
  • Roadway Design: Denser street networks (higher facility miles per square mile) are strongly associated with lower emissions, suggesting that walkable neighbourhoods with shorter block lengths reduce traffic volume concentration.
  • Spatial Spillovers: The study finds strong scale dependence. Neighbouring land use diversity has a larger positive effect on hosted emissions than local diversity, indicating that transport emissions operate within a broader commuting radius (10 km) rather than just shared borders.
  • The "Hosting" Effect: Neighbourhoods with desirable urban form features (density, mixed uses) tend to host the transportation infrastructure (arterial roads) that generates regional mobility. Consequently, these areas bear higher localized production-based emissions, even if their residents generate less travel demand. This highlights an environmental justice dimension where urban cores bear air quality impacts for suburban travel demand.
• Residential Emissions:
  • Density: Higher local population density is consistently associated with lower per-capita residential emissions for both electricity (-6.8%) and non-electricity energy (-12.0%). This aligns with the mechanism that denser housing implies smaller units, shared walls, and lower energy demand.
  • Metropolitan Scale: Regional (CBSA) density and diversity effects are statistically insignificant in the joint models, suggesting that local design and retrofit measures are more critical than broad regional density targets for residential emissions.
  • Diversity: Land use diversity shows a modest positive association with residential electricity emissions but limited influence on non-electricity energy.

Significance and Claims
The paper argues that both macro and micro perspectives are necessary but insufficient in isolation for effective urban climate policy. The findings support coordinated multi-scale mitigation strategies in the United States.

Specifically, the production-based perspective reveals that infrastructure carbon burdens are spatially mismatched with household travel behavior. Urban cores often "host" emissions for surrounding regions, creating a need for policies that address these externalities. The study suggests that policy interventions should target diverse 10-km commuting-shed clusters rather than isolated block-level interventions. Potential policy applications discussed include congestion pricing in urban cores, low-emission zones, revenue redistribution from infrastructure user fees to hosting communities, and the development of mobility hubs at regional peripheries.

The author notes that while the results are grounded in US-specific emissions accounting and urban form, the empirical approach is transferable to other regions with comparable gridded emissions and land use data. The study acknowledges limitations, including residual allocation bias in transportation emissions at the neighbourhood scale and the cross-sectional nature of the data (2019), which limits the ability to capture post-pandemic behavioral shifts.