More extremely hot days, more heat exposure and fewer cooling options for people of color in Connecticut, U.S.

Urban heat exposure is projected to increase with global warming, urban expansion, and population growth, leading to significant health, economic, and environmental impacts. Prior research in the United States shows that people of color are disproportionately exposed to stronger urban heat island effects and moist heat stress and often have fewer adaptation options, increasing vulnerability. However, evidence has largely come from single-year analyses at coarse spatial scales in large cities, leaving knowledge gaps about temporal changes and conditions in small and mid-sized cities, as well as fine-scale links between exposure and adaptation strategies such as tree cover and air conditioning. This study examines heat inequality in the ten largest cities in Connecticut, focusing on two dimensions: exposure (overall temperatures, extremely hot days, and temporal change in exposure) and adaptation (tree cover and air conditioning ownership rates). The research questions are: (1) How do overall temperatures differ between people of color and white populations? (2) How many extremely hot days did people of color and white people experience over time? (3) How did heat exposure change over time for people of color and white people? (4) How do tree cover and air conditioning ownership rates differ between people of color and white people?

Evidence across many US cities shows disproportionate heat exposure among people of color, including stronger surface urban heat island effects and higher moist heat stress than white populations. People of color are also more vulnerable due to lower adaptive capacity, such as reduced access to cooling strategies. Most prior studies examine one year at census tract or block-group scales and primarily focus on large cities, despite a substantial share of the US population living in small and mid-sized cities. Predominantly people of color communities tend to have disproportionately low tree cover compared to whiter areas, yet few studies assess how vegetation and canopy cover change over time within these communities. Air conditioning is a key indoor cooling strategy associated with reduced heat-related morbidity and mortality, but ownership is unequal across demographic groups. While some metropolitan-scale analyses show lower AC ownership probabilities among Black, Hispanic, or Latino communities, fine-scale empirical evidence of intra-urban AC disparities remains scarce. Overall, comprehensive, fine-resolution, multi-decadal analyses linking heat exposure, vegetation, and AC access in small and mid-sized cities are lacking, motivating this study.

Study area: The ten largest cities in Connecticut by population (Bridgeport, Stamford, New Haven, Hartford, Waterbury, Norwalk, Danbury, New Britain, Hamden, and Manchester). These cities contain many environmental justice communities and about one-third of the state population but 70% of people of color and 52% of low-income residents. Data sources and periods: United States Census block-level race data for 1990, 2010, and 2020 (downloaded via IPUMS/NHGIS). Near-surface air temperature at 1 km for 2003–2020, combining ground stations and satellites. Landsat Collection 2 land surface temperature (LST) and NDVI time series at 30 m (synthetic annual values) for 1990–2022. High-resolution NAIP airborne imagery (60 cm) from summer 2018 for tree and vegetation cover mapping. Parcel and CAMA (property assessment) data for AC availability (2022); OpenStreetMap building footprints for building-level joins. Demographic definitions: Percentage of people of color (POC) equals one minus percentage of white alone for each census block. Blocks with POC percentage greater than 50% are categorized as predominantly POC; otherwise as predominantly white. Percentages were also computed for American Indian/Alaska Native alone, Black/African American alone, Hispanic/Latino alone, Asian/Pacific Islander alone, and mixed races. Air temperature: Computed mean daily maximum summer (June–September) air temperatures at the census tract level for 2003–2020. Trend analysis used linear regressions of temperature on year. Decadal comparisons matched census years to nearby summers: 2000 analysis used 2003 air temperature with 2000 census; 2010 used mean of 2009–2010 with 2010 census; 2020 used mean of 2019–2020 with 2020 census. Extremely hot days were defined as days with air temperature above 90°F (32.2°C). Heat exposure was calculated as the number of extremely hot days times the population (separately for POC and white populations). LST time series: Landsat Collection 2 LST cleared of clouds and shadows using QA bands. The Continuous Change Detection and Classification (CCDC) model was fit in Google Earth Engine to produce annual synthetic LST for July 1 from 1990 to 2022, capturing both abrupt and gradual changes and minimizing seasonality effects. LST was aggregated to census blocks for state- and city-level analyses. AC ownership rate: CAMA and parcel data for 2022 provided AC availability at parcel level. OSM building footprints were spatially joined to parcels to infer building-level AC. Block-level AC rate equals number of buildings with AC divided by total buildings per block. 2022 is the only year with comprehensive CAMA AC data statewide. Tree and vegetation cover mapping: NAIP 2018 60 cm imagery (R, G, B, NIR) was used to derive detailed land cover via object-based classification in Earth Engine. Steps: SNIC segmentation; random forest classification using object features (mean band values, standard deviation, area, perimeter, width, height). Manual corrections in complex areas (e.g., wetlands). Aggregation produced tree canopy and vegetation cover maps at 60 cm, then summarized to blocks. Accuracy assessment used 300 random samples per city (3000 total). Tree cover overall accuracies ranged 87–93%, producer’s and user’s accuracies 80–96%. Vegetation cover overall accuracies 91–97%, producer’s and user’s 85–99%. In New Haven, 97% of street tree crowns matched field inventory within 3 m. NDVI time series: Landsat Collection 2 summer clear observations, with CCDC used to generate July 1 synthetic NDVI annually from 1990 to 2022. Aggregated to blocks and summarized for decadal comparisons and mean NDVI calculations. Statistical analyses: At census tract scale, compared mean summer air temperatures between predominantly POC and predominantly white tracts; computed correlations between POC percentage and temperature; estimated non-spatial multivariate linear models and spatial lag models with 2020 air temperature as response and covariates including POC percentage, female percentage, percentage aged 65+, and median household income. Counted extremely hot days and computed per-person counts and total heat exposure. At census block scale, compared mean LST, tree cover, vegetation cover, NDVI, and AC rates between predominantly POC and predominantly white communities at state and city levels; computed correlations between POC percentage and each variable; assessed disparities and their changes over time (1990, 2010, 2020 for LST and NDVI). Also analyzed relationships between LST/NDVI and specific racial groups’ percentages and examined NDVI change versus LST change.

• Predominantly people of color (POC) communities experienced higher summer air temperatures than predominantly white communities: 28.18°C vs 27.83°C (difference 0.35°C; p<0.001). Summer air temperature was significantly positively correlated with POC percentage (p<0.001), and multivariate and spatial lag models associated higher POC percentages with higher temperatures in 2020.
• Extremely hot days (above 90°F, 32.2°C) were more frequent in POC communities: from 2003 to 2020, predominantly POC communities experienced 35 more extremely hot days than predominantly white communities (roughly 2 more days per year; p<0.001). Per-person extremely hot days increased from 5.7 (2000) to 11.0 (2020) for POC, versus 5.1 to 9.5 for white populations. Heat exposure (extremely hot days times population) rose more for POC due to both temperature increases and population growth.
• Summer air temperature trends: 2003–2020 annual increases were higher in predominantly POC communities (0.0749°C per year) than in predominantly white communities (0.0698°C per year).
• Land surface temperature (LST) disparities were substantial: across all blocks in the ten cities, mean LST was 4°C higher in predominantly POC communities (37°C) than in predominantly white communities (33°C; p<0.001). By city, the mean LST disparity ranged from 1.1°C to 5.1°C, increasing over time in seven cities (1990–2020). For nine of ten cities, LST in 1990, 2010, and 2020 was positively correlated with POC percentage.
• Racial group correlations: Percent mixed races showed stronger relationships with LST than any single non-white group, though slightly weaker than overall POC percentage. Among single groups, Black or African American percentage had the strongest positive correlation with LST in all years.
• Air conditioning (AC) ownership rates were lower in hotter and more POC areas. In 2022, mean AC rate in predominantly POC communities was 23% lower than in predominantly white communities (p<0.001). City-level disparities ranged from 4% to 29%, with eight of ten cities showing disparities of at least 15%. AC rate was negatively correlated with POC percentage in all cities and significant in nine of ten (p<0.01). Blocks above the statewide median LST had a mean AC rate of 22% versus 36% in cooler blocks (difference 14%; p<0.001), and AC rate was negatively correlated with LST.
• Tree canopy and vegetation were lower in POC communities. Mean tree cover was 39% in predominantly white communities versus 24% in predominantly POC communities (15% lower; p<0.001). City-level disparities in tree cover ranged from 2% to 22% and were significant in nine of ten cities; POC percentage was negatively correlated with tree cover in nine of ten cities. NDVI was consistently lower in predominantly POC communities in 1990, 2010, and 2020, with negative correlations in nine of ten cities and disparities increasing in about half of the cities.
• Vegetation change mitigated but did not offset warming. NDVI change was significantly and negatively correlated with LST change in all cities, indicating that increasing vegetation helps reduce temperature increases; however, most blocks still experienced increases in both LST and NDVI, suggesting the vegetation increase to date has been insufficient to fully counteract warming.
• Vulnerability nexus: Blocks with high POC percentages tended to have high LST and simultaneously lower AC access and lower tree cover, compounding heat risk.

The findings directly address the research questions by demonstrating that people of color in Connecticut’s ten largest cities face higher overall temperatures, more extremely hot days, and larger increases in heat exposure over time, while simultaneously having fewer adaptation options through lower AC ownership and tree canopy. These disparities persisted or widened across three decades for surface temperatures and vegetation metrics, and are strongest for communities with higher Black or African American or mixed-race percentages. Policy implications include prioritizing heat adaptation in underserved communities. Tree planting and AC provide complementary benefits with different spatial and temporal profiles. Trees offer neighborhood-scale cooling, long-term and increasing benefits, and additional ecosystem services, although they take years to mature. AC provides immediate indoor relief but may add outdoor heat and lacks co-benefits. Given rising urban temperatures and more frequent extreme heat, tree-planting initiatives should prioritize communities with high temperatures, low tree cover, and low AC rates, with strong community engagement for planting, maintenance, and stewardship. Equity and environmental justice should be central in defining priority areas, and future tools should explicitly account for race and incorporate heat exposure and vulnerability dimensions, which current screening tools often omit. The study’s multi-temporal, fine-scale geospatial products can guide ongoing policy efforts in Connecticut to increase urban canopy in large cities and environmental justice communities. Broader access to fine-scale AC datasets would significantly improve heat vulnerability identification and decision-making.

This study integrates multi-decadal satellite observations, high-resolution airborne imagery, census data, parcels and property assessments, and OSM to quantify disparities in urban heat exposure and adaptation capacity across ten Connecticut cities. It shows that predominantly people of color communities are hotter, experience more extremely hot days, and are warming faster over time, while having lower AC ownership and less tree canopy. Vegetation increases mitigate heat but are insufficient to offset overall warming to date. The results underscore the urgency of targeted, equitable interventions such as prioritizing tree planting in communities with high temperatures and low AC access, and improving access to cooling options. The geospatial datasets produced can support policy implementation and planning. Future research should: (1) evaluate additional indoor and outdoor adaptation strategies beyond AC and tree canopy; (2) assess AC usage patterns and barriers, not only ownership; (3) quantify canopy expansion potential and species selection to maximize cooling and co-benefits; (4) consider humidity and wet-bulb impacts of greening in different climate zones; (5) incorporate mobility and occupational exposure to better estimate individual heat exposure; and (6) extend analyses to other small and mid-sized US cities to assess generalizability.

Key limitations include: (1) Adaptation capacity was proxied primarily by AC ownership; other strategies were not analyzed due to data constraints. AC ownership does not equal usage, which may vary by cost and behavior. (2) AC availability data from CAMA may underestimate window units, especially in renter-dominated buildings, and was available only for 2022. (3) Mobility and workplace conditions were not incorporated; exposures differ between indoor and outdoor workers and by commute patterns. (4) The study was limited to ten Connecticut cities; findings may not generalize to other regions or city sizes. (5) Fine-scale AC datasets are limited in availability and openness, constraining broader comparisons.