Historical red-lining is associated with fossil fuel power plant siting and present-day inequalities in air pollutant emissions

Ambient air pollution causes over 100,000 premature deaths annually in the United States, with Black, Hispanic and Asian communities facing higher exposures and proximity to polluting industries even after accounting for socioeconomic factors. Fossil fuel power plants emit particulate matter (PM), nitrogen oxides (NOx), and, for coal, sulfur oxides and mercury, contributing to unequal pollution burdens that have historically been higher for people of colour. During the 1930s, the Home Owners' Loan Corporation (HOLC) graded neighbourhoods by perceived mortgage lending risk, codifying racist beliefs that led to 'red-lining' and deepened segregation. This study asks whether HOLC grades influenced subsequent siting of fossil fuel power plants and whether those historical siting patterns contribute to present-day inequalities in power plant emissions burdens (NOx, SO2, PM2.5). The primary objective is to test whether worse HOLC grades (especially D versus C) are associated with a higher likelihood of new plant siting upwind and within 5 km over successive periods since the 1940s; the secondary objective is to assess whether these historical siting differences translate into higher present-day emissions burdens for worse-graded neighbourhoods.

Digitization of HOLC maps has enabled research linking historical red-lining to current environmental and health disparities. Prior studies have associated worse HOLC grades with less urban vegetation, higher surface temperatures, higher ambient air pollution, the siting of oil and gas wells, and adverse health outcomes including preterm birth, asthma emergency visits, and late-stage cancer diagnoses. However, most analyses have been cross-sectional due to limited historical outcome data, offering limited insight into trajectories over time or control for pre-existing neighbourhood differences. Broader environmental justice literature shows communities of colour disproportionately bear industrial siting and emissions burdens; reductions from coal retirements have benefited White Americans more. There is evidence that demographic and socioeconomic characteristics, and anticipated community resistance, influence facility siting decisions. This study extends prior work by examining longitudinal plant siting across decades using HOLC grades and linking to present-day emissions from fossil fuel power plants, while attempting to account for pre-existing plant presence and baseline socioeconomic factors.

- Study design and units: HOLC neighbourhood polygons from Mapping Inequality were the analysis units. Neighbourhoods were considered 'exposed' if located downwind of a fossil fuel power plant within 5 km of the HOLC boundary, using predominant wind direction. - Power plant data: Plant- and generator-level data from EIA Form 860 (2019) were used to construct a dataset of 3,284 fossil fuel power plants (continental US). Plants in Alaska, Hawaii, Puerto Rico were excluded. Plant first operation date, retirement, primary energy source (coal, petroleum, natural gas) and nameplate capacity were derived from generators. Plants with nuclear or renewable primary sources were excluded except a small number of pre-1940 biomass plants for baseline characterization. - Plant categories: Analyses considered all fossil fuel plants and separately coal- or oil-fired plants, and peaker plants (plants primarily burning oil and combustion turbines primarily burning natural gas, identified via EIA Data Atlas). - Periods of siting: Siting was assessed for plants first operational pre-1940 (baseline), 1940–1969, 1970–1999, and 2000–2019. All plants first online during a period were included irrespective of later retirement. - Wind and exposure definition: Predominant wind direction at plant locations was determined from NLDAS-2 monthly winds (u, v) for 1980 (midpoint of study period; highly consistent with 2020), categorized into N/E/S/W via atan2(v/u). Around each plant, a 5 km buffer was sliced into four 90° sectors; the downwind sector defined exposed HOLC polygons via intersection. Sensitivity analyses used proximity irrespective of wind and a 10 km distance. - Emissions burden: For present-day analysis, emissions of NOx (2019), SO2 (2019) from EPA eGRID, and PM2.5 (2018 draft eGRID) were summed across all upwind plants within 5 km intersecting each HOLC neighbourhood boundary. Emissions reflect plants still operable in 2019 (N=2,878 matched plants). Unadjusted eGRID emissions were used. - Covariates: All models controlled for presence of an upwind plant within 5 km before 1940 and US Census region (Northeast, Midwest, South, West). Sensitivity models additionally controlled for baseline (1940) city population and HOLC area description sheet socioeconomic variables (percent foreign-born, percent Black, median household income, median building age, and ordinal indices for occupation, property repair status, and mortgage availability) where available. - Comparative strategy: To reduce confounding by large baseline differences, adjacent HOLC grade comparisons were used (D vs C; C vs B; B vs A). - Statistical analysis: For siting (presence within 5 km), Poisson regression with robust standard errors estimated prevalence ratios (PRs), stratified by period. For counts of nearby plants, negative binomial models estimated incidence rate ratios (IRRs), stratified by period. For emissions burden, log-linear regressions estimated geometric mean ratios (GMRs) after log-transforming pollutant totals. Emissions models were restricted to neighbourhoods with at least one upwind plant within 5 km still operable in 2019. Sensitivity analyses used proximity-only exposure and a 10 km buffer. Spatial processing used NAD 1983 Contiguous USA Albers; analyses in R 4.2.0.

- Sample: 8,871 HOLC neighbourhoods in 196 urban areas; 3,284 fossil fuel power plants. A slightly higher proportion of D-graded neighbourhoods already had an upwind plant within 5 km before 1940 (1.9%) vs C (1.1%), B (1.2%), A (0.9%). Disparities increased over time, with greater proportions and numbers of upwind plants near worse grades. - Siting (D vs C), adjusted for pre-1940 plants and region: • Presence of any fossil fuel plant within 5 km upwind: PR (95% CI) = 1.72 (1.49, 1.98) for 1940–1969 (+72% risk), 1.20 (1.07, 1.35) for 1970–1999 (+20%), 1.31 (1.14, 1.52) for 2000–2019 (+31%). • Coal or oil plants: approximately +86%, +22%, +3% higher risk across the three periods (strongest earlier, minimal in 2000–2019). • Peaker plants: approximately +133%, +33%, +53% higher risk across the three periods. • Number of upwind plants within 5 km: IRR (95% CI) = 1.82 (1.56, 2.12) for 1940–1969, 1.31 (1.14, 1.50) for 1970–1999, 1.42 (1.19, 1.69) for 2000–2019. - Adjacent comparisons C vs B and B vs A also showed increased risks, generally smaller and less precise than D vs C and less consistent across periods. - Sensitivity and robustness: • Adjusting for 1940 city population and HOLC socioeconomic variables yielded similar or stronger D vs C associations (except weaker during 1970–1999), with wider CIs due to reduced sample size. • Using 10 km instead of 5 km attenuated associations; for D vs C, siting associations remained apparent primarily in 1940–1969. • Defining exposure by proximity only (ignoring wind) produced generally consistent but weaker estimates. - Present-day emissions burden (among neighbourhoods with at least one upwind plant within 5 km still operable in 2019): clear grade gradient (A < B < C < D), especially for NOx and PM2.5. • D vs C GMRs (95% CI), adjusted for pre-1940 plants and region: NOx 1.82 (1.44, 2.31) (+82%); SO2 1.38 (1.17, 1.63) (+38%); PM2.5 1.63 (1.37, 1.94) (+63%). Smaller positive associations for C vs B; minimal for B vs A. • Additional adjustment for baseline population and socioeconomics: similar patterns with reduced precision; sensitivity analyses suggested slightly stronger D vs C associations (e.g., NOx GMR 1.98 (1.32, 2.98); SO2 1.46 (1.09, 1.96); PM2.5 1.86 (1.38, 2.52)). • Using 10 km attenuated effect sizes but patterns persisted.

The study shows that HOLC red-lining grades from the 1930s are associated with the subsequent siting of fossil fuel power plants and with higher current emissions burdens near worse-graded neighbourhoods. The strongest and most consistent disparities appear when comparing D ('hazardous') versus C ('declining') areas, and associations persist into 2000–2019, over seven decades after map creation. These findings support the hypothesis that structural racism encoded in housing market appraisals helped shape environmental hazard siting patterns, contributing to present-day inequality in emissions burden. While the analysis does not identify specific causal mechanisms, plausible pathways include depressed land values, reduced homeownership and incomes, deepened segregation, disinvestment, and diminished political power in red-lined communities that reduced their ability to resist locally unwanted land uses or made industrial employment more attractive. Long plant lifetimes (median retirement ~49 years) and generally higher emissions rates from older and peaker units likely help maintain present disparities. Broader energy transitions and regulatory changes have reduced power sector emissions and narrowed some ambient PM2.5 disparities since 2000, but inequities remain. Findings align with prior national studies linking HOLC grades to higher ambient air toxics risks and NO2/PM2.5 levels but provide source-specific evidence for fossil fuel power plants and focus on emissions rather than ambient concentrations. The results underscore the enduring public health relevance of historical housing discrimination for environmental justice and can inform policies targeting equitable energy and air quality improvements.

Racism codified in historic HOLC red-lining maps contributed to higher likelihood of nearby fossil fuel power plant siting and higher present-day emissions burdens (NOx, SO2, PM2.5) in worse-graded neighbourhoods, particularly D versus C. These disparities persisted across multiple siting eras up to 2019 and reflect both siting decisions and characteristics of plants (fuel, technology, controls). The study’s longitudinal, national analysis advances understanding of how structural racism shaped the geography of electricity generation and current emissions burdens. Future research should elucidate city- and region-specific mechanisms (e.g., land values, zoning, political processes, industrial co-location), incorporate additional historical industrial data, assess other infrastructures (e.g., highways), and directly link emissions burdens to ambient exposures and health outcomes to quantify absolute health impacts and guide targeted mitigation.

- Causality: Inability to fully isolate causal effects of red-lining due to potential unmeasured confounders and pre-existing differences; although models controlled for pre-1940 plant presence, region, and in sensitivity analyses baseline population and socioeconomic factors. - Historical industrial context: Lack of data on other industries or industrial zoning in the 1930s that may correlate with both worse HOLC grades and later power plant siting. - Data limitations: Missing and incomplete data in EIA and EPA sources could misclassify plant characteristics and underestimate emissions; emissions available only for plants operable in 2019; PM2.5 estimates from draft eGRID (2018). - Modeling assumptions: Highly skewed emissions with residual non-normality even after log transformation; wind direction based on 1980 conditions used for all periods; exposure defined by distance buffers and predominant wind without dispersion modeling. - Scope: Emissions burden used as a proxy for potential impact without translating to ambient concentrations or health effects; results may not capture downwind impacts beyond 5–10 km; other infrastructures (e.g., highways) and community demographic changes over time were not assessed. - Generalizability across grades: Strongest and most consistent associations were for D vs C; comparisons for other adjacent grades were smaller and less precise.