Racial/ethnic disparities in the distribution of heatwave frequency and expected economic losses in the US

Previous research on social disparities in heat exposure has not examined heatwave frequency or economic damage at the local or neighborhood level. Additionally, most US studies have focused on specific cities or regions, and few national-scale studies encompassing both urban and rural areas have been conducted. These gaps are addressed here by analyzing racial/ethnic disparities in the distribution of annual heatwave frequency and expected economic losses from heatwave occurrence in the contiguous US. Census tract-level data on annualized heatwave frequency and expected loss from the FEMA’s National Risk Index are linked to relevant variables from the American Community Survey. Results indicate that all racial/ethnic minority groups except non-Hispanic Black are significantly overrepresented in neighborhoods with greater annual heatwave frequency (top 10% nationally), and all minority groups are overrepresented in neighborhoods with greater total expected annual loss from heatwaves, compared to non-Hispanic Whites. Multivariable models that control for spatial clustering, climate zone, and relevant socio-demographic factors reveal similar racial/ethnic disparities, and suggest significantly greater heatwave frequency and economic losses in neighborhoods with higher percentages of Hispanics and American Indians. These findings represent an important starting point for more detailed investigations on the adverse impacts of heatwaves for US minority populations and formulating appropriate policy interventions.

www.nature.com/scientificreports/(as defined by the area's local weather forecast office) outside the historical averages.Modeled estimates of annualized heatwave frequency and total expected annual loss (combines and quantifies losses for three consequence types: agriculture, building, and human fatalities/injuries) are linked to relevant variables from the American Community Survey (ACS) five-year estimates.

Results
Figure 1 depicts the distribution of census tracts in the top 10% (at or above the 90th percentile) for annualized heatwave frequency and total expected loss from heatwave occurrence, respectively, in the contiguous US.These tracts are concentrated mainly in southwestern states (e.g., Arizona and California), as well as those in the lower Midwest (e.g., Kansas and Missouri).Figure 2 shows how relationships between annualized heatwave frequency and specific racial/ethnic minority groups vary spatially across the US.These bivariate maps suggest that tracts in the highest tercile (top 33.3%) for both variables are located mainly in the West (e.g., California) for Hispanic percentage and the South (e.g., Arkansas, Louisiana, and Mississippi) for non-Hispanic Black percentage.Tracts in the highest tercile for both heatwave frequency and American Indian percentage indicate a more dispersed pattern, and can be found in the West, Midwest, and South (e.g., California, Nevada, Kansas, and Oklahoma).The spatial distribution of relationships between total expected annual loss from heatwaves and racial/ethnic minority groups are shown in Fig. 3. Tracts in the highest tercile (top 33.3%) for both variables can be found in multiple states of the West for Hispanic percentage and South for non-Hispanic Black percentage.Tracts in highest tercile for both total expected loss and American Indian percentage again suggest a more dispersed pattern and are located in multiple US regions (e.g., West, Midwest, and South).
Statistical results from racial/ethnic comparisons are summarized in Table 1, where each minority group's percentage (based on their national total) in the top decile for each heat indicator is compared to the corresponding non-Hispanic White percentage.For tracts in the top 10% for heatwave frequency (Figure 1a), risk ratios (RRs) significantly exceed 1.0 (p < 0.001) for all minority groups, except non-Hispanic Black.Specifically, Hispanic, American Indian, and Asian persons are 71.12%,105.4% and 45.4% more likely, respectively, to reside in these top 10% tracts, while Black residents are 5.7% less likely (compared to non-Hispanic Whites).For tracts in the top 10% nationally for total expected annual loss (Figure 1b), RRs for all minority groups significantly exceed 1.0 (p < 0.001).Specifically, Hispanic, non-Hispanic Black, American Indian, and Asian persons are 16.5%, 34.9%, 49.1%, and 3.6% more likely, respectively, to reside in these tracts.American Indians indicate the greatest overrepresentation (highest RRs) with respect to non-Hispanic Whites within tracts in the top 10% for both heatwave frequency and total expected loss.
Results from multivariable generalized estimating equation models for predicting annualized heatwave frequency and total expected annual loss, respectively, based on racial/ethnic characteristics and a set of control variables are presented in Table 2.After accounting for spatial clustering of tracts, relevant socio-demographic factors known to be associated with extreme heat exposure in the US 7,10 , and climate zones defined by the U.S. Department of Energy, annualized heatwave frequency indicates a statistically significant and positive relationship (p < 0.05) with the percentages of Hispanic and American Indian residents.All racial/ethnic groups reveal significantly positive associations (p < 0.05) with total expected annual loss, with the exception of non-Hispanic Asians.The percentages of Hispanics and American Indians represent the only racial/ethnic variables to yield a positive and significant (p < 0.01) coefficient in both models.
Expected annual losses associated with the three consequence types that comprise total expected annual loss were also analyzed separately.Results from multivariable models focusing on expected losses related to agriculture, buildings, and population equivalence (inflation-adjusted dollar value assigned to the sum of human fatalities and injuries), respectively, are summarized in Table 3 (control variables not shown).Expected annual loss for agriculture reveals a significantly positive association (p < 0.001) with the percentages of Hispanic and American Indian residents, but a negative association (p < 0.001) with other racial/ethnic variables.Expected annual loss related to buildings indicates a significantly negative (p < 0.001) or non-significant association (p > 0.05) with Table 1.Distribution of persons by race/ethnicity in tracts in the top 10% for National Risk Index (NRI) modeled heatwave annualized frequency (event-days per year) and total expected annual loss from heatwave occurrence (US dollars).N = 83,481 census tracts in contiguous US.Risk ratios (RRs) are based on dividing minority group percentages (based on their total population in continental US) by corresponding NH White percentage.

Tracts in top 10% of heatwave annualized frequency
Tracts in top 10% of total expected annual loss from heatwave occurrence

Percent (%) Risk ratio 95% CI of risk ratio Z-test statistic (p value) Percent (%) Risk ratio 95% CI of risk ratio Z-test statistic (p value)
Hispanic (of any race)

Discussion
This study extends thermal inequity research by conducting a national-scale analysis of racial/ethnic disparities in local heatwave frequency and expected economic loss from heatwave occurrence-indicators that have not been examined previously.The first research question sought to determine whether racial/ethnic minority groups are overrepresented in neighborhoods with the highest heatwave frequency and total expected loss from heatwaves (i.e., tracts in the top 10% nationally), compared to the non-Hispanic White population.Results indicate that all racial/ethnic minority groups except non-Hispanic Black are significantly overrepresented in neighborhoods with the highest heatwave frequency (top 10%), and all minority groups examined are overrepresented in neighborhoods with the highest total expected loss.
The second research question focused on analyzing racial/ethnic disparities in the distribution of annualized heatwave frequency and expected loss in the contiguous US, after controlling for the effects of socioeconomic factors, younger and older age, population density, metropolitan status, climate zone, and spatial clustering by US county.Multivariable models indicated increased heatwave frequency in neighborhoods with higher percentages of Hispanic and American Indian residents.Expected economic loss from heatwaves is found to increase significantly in neighborhoods with higher percentages of Hispanic, Black, American Indian, and other non-Hispanic minorities who are not Asian.Thus, neighborhoods characterized by greater American Indian prevalence are disproportionately impacted by both higher heatwave frequency and economic damage.When total economic loss from heatwave occurrence was disaggregated by consequence type, racial/ethnic disparities were found to Table 2. Multivariable generalized estimating equations for predicting heatwave annualized frequency and total expected annual loss from heatwave occurrence.N = 82,747 tracts in contiguous US with at least 500 persons.Both models use an independent correlation matrix and Tweedie (index parameter = 1.5) distribution with logarithmic link function, and adjust for clustering based on the county in which tract is located (3101 clusters).Numbers in the 'p value' column represent two-tailed p values from the Wald Chi-Square test.All continuous independent variables were standardized before model entry.Reference categories include '% Non-Hispanic White' for racial/ethnic groups and 'Very cold/cold' climate zones for the U.S. Department of Energy climate zone designation.global multivariable model to analyze statistical associations for this national scale analysis can mask regional or local variations in relationships between the dependent and independent variables.Future research should employ spatial analytic approaches that can be used to explore spatial non-stationarity of model parameters, and identify hotspots where specific minority groups are exposed to elevated levels of heatwave occurrence and economic losses.Second, this analysis did not consider heat mitigation variables (e.g., air conditioning or tree shade) due to limited availability of reliable nationwide data.Third, the results are limited to tract-level associations because FEMA's NRI data on heatwave exposure and expected losses are unavailable for smaller spatial units.Since heatwave frequency and economic damage are disproportionately distributed with respect to racial/ ethnic minority groups at the neighborhood level, more individual and household level analyses that combine quantitative and qualitative data 22,23 are recommended to examine the adverse consequences of heatwaves exposure and formulate appropriate policy interventions that focus on protecting the health and safety of US minority populations, and American Indian communities, in particular.

Methods
This study was based on publicly available and aggregated datasets described below, and no humans or living organisms were involved.

Datasets
Data associated with the two heatwave variables were obtained from the FEMA's NRI [https:// hazar ds.fema.gov/ nri/ data-resou rces# csvDo wnload], a dataset and tool for identifying communities at risk to 18 natural hazards 17 .
Modeled census tract-level values of: (1) annualized heatwave frequency; and (2) total expected annual loss (EAL) from the latest NRI (March 2023); are used as dependent variables.In the NRI, annualized frequency of a natural hazard is defined as the expected frequency or probability of a hazard occurrence per year.The EAL for any natural hazard is calculated on the basis of a multiplicative equation that considers the consequence risk factors of natural hazard exposure, historic loss estimates, and annualized frequency of the hazard.Annualized heatwave frequency in an area is estimated using the number of recorded heatwave occurrences, in event-days, each year over a 16.9 year period of record (11/12/2005-10/06/2022).Total EAL due to heatwave occurrences during the same period are quantified by summing losses associated with agricultural value, building value, and population equivalence (inflation-adjusted dollar value assigned to sum of human fatalities and injuries).Annual data on heatwave frequency and EAL are not available in the NRI for every year in the record period; only aggregated values for the entire timeframe are provided.The statistical analysis also utilized additional dependent variables representing the EAL for each consequence type: agriculture, buildings, and population equivalence.The FEMA's NRI Technical Documentation 24 provides more information on data sources and methods for tract-level estimation of annualized heatwave frequency and EAL scores.
Data on independent variables representing socio-demographic factors were derived from the 2021 American Community Survey (ACS) five-year estimates, the most recent ACS at the time this study was conducted [https:// data.census.gov/ table?d=ACS 5-Year Estimates Detailed Tables].For race/ethnicity, the analysis included variables for the percentage of individuals identified as Hispanic (of any race) and each of the following non-Hispanic groups: White, Black, American Indian, Asian, and multi/other race.Non-Hispanic Whites were included in the bivariate comparisons but excluded from multivariate models to allow results for minority racial/ethnic groups to be interpreted relative to the non-Hispanic Whites.Socio-demographic factors linked to extreme heat exposure in previous US thermal inequity studies 3,7,10 were included as control variables in multivariate models.These included the percentages of low-income (income less than or equal to twice the federal poverty level), less than high school education, and unemployed individuals, limited English speaking households (all members 14 years or older have some difficulty with English), civilian non-institutionalized persons with a disability, those under the age of 5, and those aged 65 or more years, population density, and if the tract was located in a metropolitan area (1/0).In addition to socio-demographic factors, each model included dichotomous variables (1/0) representing climate zone designations (Hot-Dry, Hot-Humid, Marine, Mixed-Dry, and Mixed-Humid) developed by the US Department of Energy Building program 25 , with the Very Cold or Cold zones combined to serve as the reference group for comparison with other zones.

Statistical analysis
The analysis for the first research question examines statistical overrepresentation of racial/ethnic minority groups within tracts in the highest decile (top 10% nationally) for both heatwave variables.Comparisons are based on risk ratios (RRs), calculated by dividing the percentage of each minority group (based on their national total) by the corresponding percentage of non-Hispanic Whites.Two-sample z-tests for proportions are used to determine if any minority group percentage differs significantly from the non-Hispanic White percentage.
For the second research question, generalized estimating equations (GEEs) are utilized for multivariable modeling of tract-level associations between relevant heatwave indicators and the entire set of independent variables.By examining the effects of all explanatory factors simultaneously in the same multivariable model, we can identify the extent to which each racial/ethnic variable exerts independent effects on annual heatwave frequency and economic losses, after the impacts of other relevant socio-demographic factors and climate zones are considered.GEEs extend the generalized linear model to account for clustered data 26 and relax several assumptions of traditional regression (i.e., normality).Both GEE models used here control for clustering based on the US county in which the tract is located (3101 counties).The Tweedie distribution with logarithmic link function and an independent correlation matrix were chosen for these GEEs, since these model specifications provided the best fit based on the quasi-likelihood under the independence model criterion 27 .The statistical significance of variables were based on two-tailed p values from Wald's Chi-Square test.All continuous independent variables

Figure 1 .
Figure 1.Census tracts in the 90th percentile or above (top 10%) in the contiguous US by: (a) National Risk Index (NRI) modeled heatwave annualized frequency; and (b) total expected annual loss from heatwave occurrence, 2022.

Figure 2 .
Figure 2. Relationship between National Risk Index (NRI) modeled heatwave annualized frequency and major racial/ethnic minority groups by census tract.

Figure 3 .
Figure 3. Relationship between NRI modeled total expected annual loss from heatwave occurrence and major racial/ethnic minority groups by census tract.