Behavior-encoded models reveal differentiated access to public cooling environment by race and income

Extreme heat reduces productivity, worsens health outcomes, and causes substantial mortality in the United States, with non-Hispanic Black Americans facing higher mortality risks. As climate change increases the frequency, duration, and intensity of heat extremes, adaptation strategies that provide immediate protection are critical. While long-term mitigation focuses on transforming energy systems, short-term adaptation via cooling environments (indoor and outdoor) can reduce heat exposure and mortality risk, even with brief stays in cooler spaces. Public agencies increasingly deploy cooling centers through libraries, schools, and community organizations, but coverage is limited and usage patterns can favor lower-risk populations. Broader inequities in access to public spaces and services by race/ethnicity and income are well documented and may extend to cooling environments. This study defines public cooling environments (PCEs) as public open spaces or facilities that provide cooling, categorizing them as: (1) tree-covered green spaces (TCGS), which have low sensory thresholds and can be several degrees cooler than sun-exposed areas; (2) non-expense-required public environments (NERPE), such as workshops, libraries, and markets that require no cost to enter or stay; and (3) expense-required public environments (ERPE), such as cafes and dining spaces that typically require purchase. The study examines spatial distribution and differentiated accessibility to these PCE types across 40 U.S. counties, accounting for behavioral factors such as walking versus driving and vehicle ownership. It further explores intra-racial/ethnic disparities by income.

Prior work shows cooling centers can reduce heat impacts but are not widespread, and lower-risk individuals may use them more than higher-risk individuals. Inequities in access to public spaces and services, including healthcare, are prevalent across racial/ethnic and income groups and contribute to worse health outcomes (e.g., higher cancer mortality among Black people due to restricted access). Urban environmental studies indicate that green spaces’ cooling effectiveness depends on canopy cover; sun-exposed green areas are less effective, while tree canopy can reduce temperatures by several degrees. Accessibility methods such as modified two-step floating catchment area (M2SFCA) have been applied to healthcare and public facility access and are suitable for incorporating supply, demand, and travel impedance. Transportation and accessibility research further highlights differences by mode (walking vs. driving), and the role of vehicle availability and socioeconomic status in shaping access. This study builds on these strands by integrating behavior-encoded accessibility modeling to analyze disparities in access to PCEs by race/ethnicity and income.

Study area: 40 U.S. counties across diverse regions and climate divisions with high exposure to extreme heat were selected. The analysis unit is the census block group (CBG). PCE categorization and supply: PCEs were categorized as TCGS (tree canopy within CBGs), NERPE (public facilities with no expense required), and ERPE (public environments requiring expense). Supply capacity was measured as tree canopy area for TCGS and floor area for NERPE/ERPE (sourced from OpenStreetMap for facilities and areas; MRLC/USFS for canopy cover). Accessibility framework: A behavior-encoded modified two-step floating catchment area (M2SFCA) model was applied separately for walking and driving. Travel times from CBG centroids to PCE locations were computed on walking and driving road networks. A 30-minute time threshold defined catchments. A Gaussian distance-decay weight reduced weights toward 0.2 at 30 minutes (sigma set to 125). Step 1 computed PCE-specific supply-to-demand ratios within the catchment (capacity weighted by decay over the sum of demand, approximated by CBG population). Step 2 summed these ratios to each demand CBG with distance-decay to produce PCE accessibility scores for each CBG and mode. Behavior encoding: Using Census vehicle availability, populations with at least one vehicle were treated as accessing PCEs by driving; those without vehicles were treated as walking. Behavior-weighted accessibility was computed as a weighted combination of walking and driving accessibility using the respective population shares within each CBG. Racial/ethnic analysis: Overrepresented CBGs for each race/ethnicity (Hispanic or Latino, non-Hispanic White, Black, American Indian, Asian) were identified by applying a threshold equal to the county-level average proportion for each group. Mean PCE accessibility for overrepresented CBGs was computed per group. Robustness was assessed via population-weighted accessibility across all CBGs. Income-stratified analysis: Within each race/ethnicity, regression analyses related PCE accessibility to income (using log-transformed accessibility to reduce extreme-value influence). Slope and significance (P-values) were reported to assess income-accessibility relationships for each PCE type. Data sources: US Census Bureau (2020 TIGER/Line boundaries, CBG-level demographics, income, vehicle availability); NOAA climate division air temperatures (2014–2023 summer averages); MRLC/USFS tree canopy cover; OpenStreetMap public facilities (locations and areas) and road networks. Analysis used Python 3.9 (pandas, numpy, matplotlib, sklearn, statsmodels). Code available at https://github.com/CLI-lhub/Behavior-encoded-Models-reveal-inequity.

- Across 40 counties, White people consistently have higher PCE accessibility than other racial/ethnic groups, especially for TCGS. - Walking scenario (overrepresented CBGs; Harris County example generalized across all 40 counties): mean accessibility for White vs. Black people: TCGS 98.51 vs. 26.64; NERPE 3.81 vs. 2.95; ERPE 0.17 vs. 0.10. White advantages: +259.7% (TCGS), +64.6% (NERPE), +28.8% (ERPE). Differences were widespread and statistically significant (Mann–Whitney U tests). - Driving scenario: disparities lessen. White accessibility is 25.8% higher (TCGS) and 8.6% higher (ERPE) than Black people, but 10.3% lower for NERPE. Spatial clustering patterns differ by race and PCE type across county regions. - Population-weighted accessibility (all CBGs): Walking: White > Black by 8.5% (TCGS), 21.1% (NERPE), 75.5% (ERPE). Driving: White > Black by 5.7% (TCGS) and 11.3% (ERPE), but White < Black by 10.2% for NERPE. - Behavioral impacts: Driving substantially increases access across percentiles versus walking (box plots). Walking and driving access are significantly positively correlated (P < 0.001). Regression (log10): TCGS driving vs. walking slope ≈ 0.743; NERPE ≈ 0.709 (R² ≈ 0.356); ERPE ≈ 1.078 (R² ≈ 0.810). Some counties show walking outperforming driving for certain PCEs (e.g., Fresno CA, Worcester MA for TCGS; Lancaster NE, Monroe NY for NERPE; Bexar TX, Fresno CA for ERPE). - Income within racial/ethnic groups (Table 1): - TCGS: income positively and significantly associated for White (slope 0.7510, P=0.0001) and Asian (0.1071, P=0.0007); non-significant for Hispanic/Latino, Black, American Indian. - NERPE: income negatively and significantly associated for all groups; largest gap for American Indian, smallest for Asian. - ERPE: income negatively and significantly associated for American Indian (−0.6730, P<0.0001) and Asian (−0.0332, P=0.0002); non-significant for White (P=0.700), Black (P=0.100), Hispanic/Latino (P=0.900).

The study demonstrates that behavioral patterns and socioeconomic context shape access to cooling environments in ways that reinforce or mitigate racial/ethnic disparities. White populations generally enjoy greater access to TCGS and ERPE, which can confer larger thermal safety benefits, particularly in walking scenarios where proximity and canopy coverage are critical. Driving raises accessibility for all groups and narrows racial disparities relative to walking, underscoring how vehicle availability mediates access. However, reliance on driving can embed inequities where vehicle ownership and income vary geographically and demographically. The negative relationship between income and NERPE access across all races/ethnicities suggests that lower-income populations rely more on no-cost public facilities for cooling, while higher-income groups may be spatially or behaviorally less connected to these resources. For ERPE, significant negative income associations in some groups (notably American Indian and Asian) indicate complex interactions between geography, facility distribution, and economic participation in expense-required environments. These findings support the research question by quantifying differentiated access by race/ethnicity and within-group income differences, and by showing how encoding behavior (walking vs. driving) is essential for realistic assessments. The results are relevant for heat adaptation planning, emphasizing equitable siting and enhancement of accessible cooling spaces, improving pedestrian connectivity, and addressing transportation barriers for populations without vehicles.

This paper introduces a behavior-encoded M2SFCA framework to evaluate access to public cooling environments across 40 U.S. counties, revealing pronounced racial/ethnic and income-based disparities. White populations generally have higher access to TCGS and ERPE, with disparities most pronounced in walking scenarios and attenuated by driving. Income is positively associated with TCGS access for White and Asian groups, while access to NERPE declines with income across all groups; ERPE access shows group-specific negative associations with income. The study underscores the need for policy actions to eliminate disparities by prioritizing equitable placement and enhancement of tree canopy, expanding and improving NERPE facilities in underserved areas, strengthening pedestrian access, and reducing transportation barriers (e.g., via transit connections and cooling routes). Future research should refine behavior modeling (e.g., multimodal access, temporal usage patterns), incorporate microclimate and facility quality measures, and evaluate policy interventions’ effectiveness in reducing heat-vulnerability disparities.