1. Introduction
Extreme heat is the primary cause of weather-related mortality in the United States and poses a substantial public health concern (Luber and McGeehin 2008). Approximately 700 heat-related deaths occur each year in the United States according to Vaidyanathan et al. (2020), although other estimates suggest that because of underreporting this total may be much larger (Weinberger et al. 2020). Exposure to extreme heat can exacerbate preexisting health conditions, contributing to additional adverse health outcomes (Bouchama et al. 2007). While extreme heat-related morbidity and mortality can often be mitigated, differential risks and exposures elevate heat vulnerability in certain communities that ultimately bear disproportionate health burdens (Uejio et al. 2011; Johnson et al. 2012).
Vulnerability to extreme heat is linked to the social determinants of health (Raphael 2006; Braveman et al. 2011; Lehnert et al. 2020), which include socioeconomic conditions as well as disparities that influence livelihoods and the environmental quality of neighborhoods (Hansen et al. 2013). Low-income communities, communities of color, children, the elderly, and individuals with chronic health conditions or disabilities are commonly considered more vulnerable to temperature extremes (Cutter et al. 2003; Yardley et al. 2011). To mitigate extreme heat exposure, cooling centers are often established by local authorities and volunteer-based organizations to assist vulnerable populations as well as other community members requiring temporary heat relief (Fraser et al. 2017). Despite the clear importance of cooling centers, “there is limited systematic and scientific knowledge regarding the composition, operation, and effectiveness of cooling center networks” (Fraser et al. 2017, p. 1038). Therefore, additional research exploring the spatial distributions of cooling center locations and their proximity to vulnerable communities is critical as urban environments prepare for more frequent extreme heat in the future (Marsha et al. 2018; Dahl et al. 2019).
Using the “Texas Triangle” megaregion as a case study, we aim to 1) quantitatively characterize the spatial distributions of cooling center locations in the San Antonio–New Braunfels (SA), Dallas–Fort Worth–Arlington (DFW), and Houston–The Woodlands–Sugar Land (HOU) metropolitan statistical areas (MSAs); and 2) evaluate the degree to which cooling centers are accessible based on distance, transportation-related factors, and indicators of socioeconomic vulnerability. The SA, DFW, and HOU MSAs include four of the largest cities (i.e., San Antonio, Dallas, Fort Worth, and Houston) in the state that implement cooling centers as part of their local climate, public health, and/or emergency management response plans during heat waves. Examining the Texas Triangle megaregion will potentially help to identify differences between the cooling center locations in each MSA that reflect localized characteristics. For example, according to the 2021 U.S. Census American Housing Survey, 87.7% of housing units in SA were estimated to have central air conditioning while the percentage was 96.5% and 93.8% in DFW and HOU, respectively (U.S. Census Bureau 2021).
Heat extremes are also projected to worsen in the future in Texas. By 2050, the number of days with high temperatures exceeding 100°F (37.8°C) in Dallas is anticipated to increase by 30–60 days (City of Dallas 2020). Similarly, the frequency of days with temperatures over 100°F (37.8°C) in San Antonio is expected to quadruple to more than 30 days per year by 2040, while the number of nights with low temperatures exceeding 80°F (26.7°C) is projected to increase from 0.03 nights to more than 2 nights per year (Sharif 2018; City of San Antonio 2019). These climatological projections and the observed increase in extreme heat event frequency across the state (Deng et al. 2018) highlight the need for accessible cooling centers throughout the Texas Triangle, which is concerning given their relatively poor spatial coverage (Kim et al. 2021).
2. Literature review
a. Heat vulnerability
Several demographic and socioeconomic factors enhance vulnerability to extreme heat: socioeconomic status, racial and ethnic minority status, age, preexisting and chronic health conditions, and disability. Communities with higher proportions of non-White residents, greater poverty rates, and lower levels of educational attainment are often more likely to be exposed to extreme heat (Hansen et al. 2013; Hoffman et al. 2020). The increased exposure of non-White populations to higher temperatures and subsequent negative heat-related health outcomes has been linked to redlining (Hoffman et al. 2020; Li et al. 2022). However, other studies have suggested that the greater burden of extreme heat exposure is more strongly linked to factors such as age, preexisting and chronic health conditions, and socioeconomic status (Whitman et al. 1997).
As a result of structural socioeconomic disadvantages, low-income communities and communities of color often experience similar vulnerability (Hansen et al. 2013). Both communities are more likely to reside in urban environments that are prone to extreme heat and may lack the social and material resources necessary to cope (Hansen et al. 2013). For example, income can influence access to and/or the ability to run air conditioning sufficiently to cool a residence. Communities of color also experience vulnerability to extreme heat as a result of underlying health disparities. Members of racial and ethnic minority groups suffer disproportionately from respiratory and cardiovascular diseases, diabetes, and other health conditions that are linked to heat vulnerability (Betancourt et al. 2003; Bao et al. 2015). In addition, members of ethnic minority language groups and foreign-born individuals may experience greater vulnerability due to linguistic isolation (Hansen et al. 2013).
Children and elderly individuals are two additional groups vulnerable to extreme heat because of their limited capacity to protect themselves and cope with such exposures. Children tend to face an increased risk of adverse health outcomes from heat extremes because of their cognitive, physiological, and psychological differences from adults (Stanberry et al. 2018), while elderly individuals tend to have underlying chronic health conditions and diminished heat-adaptation abilities. Elderly individuals are also more likely to be socially isolated and often require assistance to perform activities that other adults typically perform themselves, increasing their vulnerability further (Luber and McGeehin 2008). Risks for heat-related illnesses are compounded for individuals with underlying chronic health conditions that are documented to reduce heat-adaptation capacity (Kenny et al. 2010). Heat mortality risk has been found to increase with higher levels of disability, though this relationship can be moderated by the level of medical care accessible to the individual (Uejio et al. 2011). However, many individuals with a disability experience social factors that also adversely impact their adaptive capacity to cope with heat exposure (Gaskin et al. 2017).
b. Cooling centers
Previous research on heat-related morbidity and mortality has indicated that spending time in a cool environment reduces the risk of detrimental health outcomes (Vandentorren et al. 2006; Bouchama et al. 2007; Ostro et al. 2010). This highlights the importance of well-designed cooling center locations that adequately serve vulnerable populations. According to Widerynski et al. (2017, p. 4), a cooling center is defined as “a location, typically an air-conditioned or cooled building that has been designated as a site to provide respite and safety during extreme heat.” Cooling center locations are often formally designated, government-owned buildings, but they can also exist informally through the use of air-conditioned commercial spaces (Berisha et al. 2017; Fraser et al. 2017). No individual organization, agency, or entity is generally responsible for sanctioning and establishing cooling centers (Nayak et al. 2019), as a large variety of stakeholders are typically involved (e.g., nonprofit organizations, offices of emergency management, and public health departments). Additionally, the number of formal cooling centers throughout the United States is dynamic because many are temporarily established in response to summer heat extremes.
Cooling centers are typically incorporated within a larger heat-related health warning system or public health response plan as a low-cost measure to enhance heat resiliency (Berisha et al. 2017; Widerynski et al. 2017). They are primarily intended to serve vulnerable populations with limited access to air conditioning and those with greater hesitancy to operate air-conditioning units due to high electric utility costs during peak energy times (Widerynski et al. 2017). Often located in libraries, community or recreation centers, senior centers, and government buildings, cooling centers are primarily established in urban areas due to the higher population densities and greater access to public transportation (Fraser et al. 2017; Nayak et al. 2019; Jagai et al. 2017). Studies analyzing rural areas have suggested that the prioritization of urban cores may elevate the heat vulnerability of rural populations due to the limited availability of cooling centers, the long travel distances to existing cooling centers, and the lack of transportation options (Fechter-Leggett et al. 2016; Jagai et al. 2017).
Spatial analyses of cooling centers have highlighted various levels of accessibility within urban environments depending largely on the mode of travel. Fraser et al. (2017) found that official cooling centers in Maricopa County in Arizona (which includes Phoenix) and Los Angeles County in California (which includes Los Angeles) were accessible via walking to 2% and 3% of households, respectively. Accessibility to air-conditioned spaces improved substantially in both counties when commercial establishments were considered. Nayak et al. (2019) also identified limited accessibility via walking but determined that all cooling centers in heat-vulnerable census tracts within four metropolitan regions in the state of New York were accessible via public transit. Importantly, Voelkel et al. (2018) highlighted how accessibility analyses are very sensitive to the presumed walking speed, as the percentage of the population in Portland, Oregon, with access to a heat refuge varied from 3.4% to 32.7% depending on walking pace. In a study that considered 25 U.S. cities, Kim et al. (2021) concluded that the cooling centers in Dallas and San Antonio provided particularly poor spatial coverage. Specifically, the cities exhibited two of the three lowest standardized percentages of population coverage, where the percentage of the population within a cooling center catchment area was standardized by the total number of cooling centers in the city (Kim et al. 2021).
In addition to the degree of accessibility based upon the spatial configuration of cooling center locations, a complementary line of research based on interviews and surveys has highlighted the characteristics of cooling center visitors as well as additional individual-level barriers to accessing cooling centers (Sampson et al. 2013; Berisha et al. 2017; Mallen et al. 2022). For example, Berisha et al. (2017) found that 39%, 33%, and 18% of cooling center users in Maricopa County were White, Hispanic, and African American, respectively. Comparisons with 2020 U.S. Census data for the county (White, 53%; Hispanic, 32%; African American, 6.7%) seem to suggest that vulnerable groups were overrepresented in terms of cooling center use and may exhibit a greater need and/or willingness to access a cooling center, which could potentially benefit from increased spatial proximity. In terms of individual impediments to visiting a cooling center, common barriers include inability to bring pets, lack of access via public transit, lack of awareness, safety concerns, and distrust of the organizing entities (Sampson et al. 2013; Mallen et al. 2022; Bedi et al. 2022).
Unfortunately, the emergent nature of cooling center locations suggests they are unlikely to be organized in an optimal spatial manner that fully mitigates the health risks among the most heat-vulnerable populations (Fraser et al. 2017; Allen et al. 2022). This has potentially contributed to the low awareness and use of cooling centers. For example, of the 322 respondents who felt too warm inside their residences in a survey of Houston adults, only 36.1% knew what cooling centers were (Hayden et al. 2017). Relatedly, in Maricopa County the majority (63%) of cooling centers remained at or below 50% of their capacity on a daily basis (Berisha et al. 2017). Several studies have suggested that cooling center locations might be improved by leveraging spatial optimization techniques and geographic information systems (Bradford et al. 2015; Widerynski et al. 2017). Therefore, this study utilizes spatial analysis techniques to assess the characteristics of the cooling center locations within the SA, DFW, and HOU MSAs. By focusing on Texas, this paper expands upon prior cooling center research, which has focused largely on New York (Nayak et al. 2017, 2019), California (Bedsworth 2009; Fraser et al. 2017), and Arizona (Berisha et al. 2017; Mallen et al. 2022).
3. Methods
The study considered the HOU, DFW, and SA MSAs within the Texas Triangle megaregion (Fig. 1). An MSA includes at least one central county containing an urbanized area of at least 50 000 inhabitants as well as additional outlying counties that meet specific work-commuting thresholds to and from the central county (U.S. Census Bureau 1994). The majority of the analysis was conducted at the census tract level (i.e., we considered all census tracts within a given MSA or all census tracts within the urban core county of a given MSA) to address the spatial variability of heat vulnerability. Census tracts are small subdivisions of a county that have an average population of 4000 individuals. They typically vary in area to maintain similar population sizes, with smaller census tracts generally located in more urbanized areas.
Location of (a) the Texas Triangle megaregion, (b) cooling centers in the DFW MSA (supplemental Table S1), (c) cooling centers in the SA MSA (Table S2), and (d) cooling centers in the HOU MSA (Table S3).
Citation: Weather, Climate, and Society 15, 4; 10.1175/WCAS-D-22-0125.1
The cooling center database for the three study MSAs was constructed in the spring of 2021 using information made available by city governments and other local organizations (e.g., news networks, newspapers, faith-based organizations) (Fig. 1). These resources describing cooling center locations unfortunately provided little additional information regarding the underlying rationale for the selection of the site to serve as a cooling center, the cost incurred, or the capacity and use of the center. The hours of operation varied between locations and across various days. They were often tied to the existing hours of operation for the facility, although extensions for cooling centers services were occasionally provided for select locations. Across the three study MSAs, the earliest opening was 0700 LT and the latest closure was 2100 LT, meaning no overnight accommodation was provided. The availability during the weekend often differed from the weekday, as several locations were regularly closed on Sundays or closed on both Saturdays and Sundays. Although the thresholds differed between the various MSAs, heat index values exceeding 100°F (37.8°C) were generally necessary for the cooling centers to open. Information on the number of cooling centers within each MSA is provided in Table 1, and complete lists are provided in Tables S1–S3 in the online supplemental material.
Descriptive statistics based upon the census tracts within each study MSA and the constructed cooling center database.
The cooling center address information was geocoded to enable a quantitative evaluation of the spatial distribution of the cooling centers using two forms of clustering analysis. First, average nearest-neighbor analysis, which compares the average observed distance between nearest neighbors with an expected average based on a random spatial distribution, was performed for each MSA to statistically determine if the locations exhibited significant clustering or dispersion overall. Second, multidistance spatial cluster analysis based on Ripley’s K function was conducted to identify the specific distances at which clustered or dispersed patterns were observed in each MSA. Third, kernel density estimation was used to produce density surfaces of the cooling center locations in each MSA to enable additional visualizations.
The second portion of the study assessed the degree to which cooling centers were accessible by calculating the distance in miles from the centroid of each individual census tract within the study MSAs to the nearest cooling center and the distance in miles from each cooling center to the nearest public transportation stop. Census tracts no more than 0.5 mi (0.8 km) from a cooling center were classified as having access via walking. The half-mile threshold was selected because of its use in previous cooling center accessibility studies (Nayak et al. 2019; Kim et al. 2021). Because it is also a widely accepted parameter for public transit catchment areas (Guerra et al. 2012), a half-mile threshold was applied to determine if a cooling center was accessible via public transportation. Additionally, driving-distance thresholds were computed to determine census tract access to cooling centers via automobile. The daily average commuting time for each MSA provided by the U.S. Census Bureau (2018) American Community Survey (ACS) (24.6, 26.8, and 27.6 min for SA, DFW, and HOU, respectively) was combined with two representative driving speeds [30 mi h−1 (48 km h−1) representing residential or city driving and 70 mi h−1 (113 km h−1) representing highway driving] to define an upper and lower bound for the driving-distance threshold. If a cooling center was within the threshold, a given census tract was considered to have access at the specified speed of travel.
To analyze how the cooling center locations were related to various indicators of vulnerability, nine socioeconomic variables at the census tract level were obtained from the 2018 ACS 5-yr estimates, including percentage of the population aged 5 years or younger, percentage of the population aged 65 years or older, percentage of the population in poverty, percentage of the population with a disability, percentage of the population with no high school diploma, percentage of the population that was African American or Black, percentage of the population that was Hispanic or Latino, percentage of households with limited English-speaking abilities, and percentage of total housing units with no vehicle available. Census tracts were utilized since they provided the smallest geographic unit for which all the indicators were available. Descriptive statistics of the socioeconomic variables are provided in Table 1. To explore if areas with significantly higher vulnerability were collocated with cooling centers, hotspot cluster analysis using the Getis-Ord Gi* statistic was performed for each indicator (Getis and Ord 1992; Ord and Getis 1995). The spatial relationships among the census tracts in each MSA were defined using fixed distance bands. Overlaying the hotspot clusters with the cooling center kernel density surfaces enabled an initial visual assessment of if higher cooling center densities overlapped vulnerable population clusters.
The relationships between vulnerability and cooling center proximity were quantitatively evaluated via regression. Regression models were estimated at the census tract level for each MSA to understand the partial effects of the nine vulnerability indicators (independent variables) on distance to the nearest cooling center (dependent variable). Initially, ordinary least squares estimation was utilized, but due to the high levels of spatial autocorrelation exhibited by the residuals, spatial regression models were ultimately used to provide more robust results. Lagrange multiplier tests were performed to determine if a spatial lag or spatial error regression model would best fit the data. The significance of the Lagrange multiplier tests indicated that spatial lag regression was most appropriate for the data. Spatial lag regression was performed at the census tract level for each MSA overall as well as for the urban core counties of each MSA (Fig. 1a). The urban core counties were Dallas County and Tarrant County, which were analyzed individually, in the DFW MSA (Fig. 1b); Bexar County in the SA MSA (Fig. 1c); and Harris County in the HOU MSA (Fig. 1d). Considering the core counties individually helped determine if the relationships between vulnerability and distance to the nearest cooling center exhibited notable differences when comparing the entire MSA with the more highly urbanized core.
4. Results and discussion
a. Spatial clustering of cooling center locations
The observed mean nearest-neighbor distance of the cooling center locations varied substantially among the three study MSAs, with SA cooling centers located more proximate to one another on average (2.8 km) relative to DFW (3.6 km) and HOU (7.8 km) (Table 2). This potentially reflects the less expansive nature of urban development within SA relative to the two larger MSAs. In terms of statistical significance, the cooling centers in SA and HOU did not display significant differences between their expected and observed mean nearest-neighbor distances. Contrastingly, the DFW locations exhibited significant spatial clustering, which might be partially attributable to the pronounced multinodal structure of the MSA and the cooling centers being spatially arranged to serve the distinct urban cores of Dallas and Fort Worth.
Cooling center average nearest-neighbor analysis for each study MSA.
The results of the multidistance spatial cluster analysis provided additional insights regarding the specific distances at which clustering and dispersion occurred. Although cooling centers within SA exhibited clustering at small (1 km) and medium (between 4 and 5 km) distances, most distances exhibited a dispersed pattern (Fig. 2a). The clustering at short distances might indicate there are redundancies in the cooling center locations, as their service areas may overlap due to their proximity. However, proximate locations could be beneficial if the cooling centers were intentionally located in areas with large heat-vulnerable populations. This possibility is explored further in section 4c. DFW (Fig. 2b) and HOU (Fig. 2c) exhibited clustering over most distances, although HOU displayed a small degree of dispersion around 30 km. The strongest clustering occurred in DFW, which aligned with the nearest-neighbor analysis. The DFW clustering peaked at 15 km, further supporting the notion that the cooling center locations reflected the underlying multinodal structure of the MSA.
Multidistance spatial cluster analysis depicting the observed and expected K values for each study MSA.
Citation: Weather, Climate, and Society 15, 4; 10.1175/WCAS-D-22-0125.1
b. Cooling center accessibility
The distance between census tract centroids and the nearest cooling center within the study MSAs ranged from 0.10 to 91.65 km. Greater distances were generally observed in the peripheral regions of the MSAs. The accessibility analysis indicated that cooling centers were only within walking distance for a small fraction of census tracts (Table 3). DFW exhibited the highest degree of accessibility via walking, with 3% of census tracts within the half-mile threshold. The percentage was lower for HOU and SA, where only 1% and 2% of tracts, respectively, were accessible to cooling centers via walking. These percentages aligned with an analysis of Maricopa and Los Angeles counties, which found that official cooling centers were only accessible to 2% and 3% of households via average walking speeds (Fraser et al. 2017). Contrastingly, Voelkel et al. (2018) determined that 16.9% of the Portland, Oregon, population was within walking distance of a cooling center. Although this discrepancy might be partially attributable to the more confined study region (i.e., city rather than county), it also may reflect the different urban development patterns within the Pacific Northwest relative to the Sun Belt.
Accessibility to cooling centers by walking, driving, and proximity to public transportation stops for each study MSA.
Texas Triangle cooling center locations were more accessible when considering proximity to public transportation stops (Table 3). In SA, all 27 cooling centers were accessible via public transportation, while a similarly high percentage of cooling centers (98%) were also accessible by a transit stop in HOU. DFW had the lowest accessibility, with only 68% of cooling centers within a half mile of a transit stop. This might be partly attributable to the higher proportion of cooling centers in DFW (5.8% vs 0% in HOU and SA) associated with volunteer-based organizations that were not located at existing public facilities. Contrastingly, the higher level of cooling center accessibility based on proximity to public transportation observed in HOU and SA is likely due to the cooling centers being located at public facilities, such as libraries, senior centers, and community centers, where public transportation networks are typically robust. Nayak et al. (2019) observed similarly high levels of accessibility in New York metropolitan regions, with 80% of cooling centers within a half mile of a public transit stop. This seems to suggest that the low level of cooling center accessibility via walking throughout the Texas Triangle relative to New York is at least partly overcome by public transit infrastructure.
The accessibility analysis indicated that cooling centers were highly accessible via driving (Table 3). DFW exhibited the highest degree of accessibility, with 87% and 98% of census tracts accessible to cooling centers based on the 30 and 70 mi h−1 distance thresholds, respectively. HOU and SA had the same degree of accessibility for the 30 mi h−1 threshold, with 81% of census tracts accessible to a cooling center, but a higher level of accessibility was observed for HOU (96%) relative to SA (93%) for the distance threshold based on the 70 mi h−1 speed of travel. Although a large proportion of census tracts were at an accessible distance from cooling centers even at slow driving speeds, it is problematic to assume that heat-vulnerable populations have access to a personal vehicle, due to the costs of owning and maintaining a car. Similarly, the high degree of cooling center accessibility due to proximity to public transit does not fully acknowledge the challenges of navigating public transit systems (e.g., waiting times, transfers, delays, and costs).
c. Cooling center density surfaces and vulnerability hotspot cluster analysis
The cooling center density surfaces within SA (Fig. 3) and DFW (Fig. 4) illustrated that the cooling centers were located primarily within the urban core counties of Bexar and Dallas/Tarrant, respectively. Although higher cooling center densities in HOU (Fig. 5) were also observed within urban core counties (i.e., Harris and Fort Bend), the density surface exhibited a greater spatial extent as it stretched southeastward toward Galveston Island and incorporated southern portions of the MSA. Collectively, the density surfaces generally highlighted a lack of cooling center coverage within the peripheral regions of the MSAs, which may elevate heat vulnerability in these outlying counties. Nayak et al. (2019) highlighted similar accessibility issues in rural regions of New York.
Spatial clustering analysis of census tracts in SA for individuals (a) aged 65 years and older, (b) aged 5 years and younger, and (c) with a disability compared with the cooling center density surface.
Citation: Weather, Climate, and Society 15, 4; 10.1175/WCAS-D-22-0125.1
As in Fig. 3, but for DFW.
Citation: Weather, Climate, and Society 15, 4; 10.1175/WCAS-D-22-0125.1
As in Fig. 3, but for HOU.
Citation: Weather, Climate, and Society 15, 4; 10.1175/WCAS-D-22-0125.1
Overlaying the density surfaces with the hotspot cluster analysis of select indicators of heat vulnerability provided an initial understanding of the degree to which the cooling center locations overlapped with the populations they typically aim to serve. Specifically, disability and age were used as examples since they also emerged as significant variables in the regression modeling. In SA, a visible disconnect was observed between the cooling center locations and individuals aged 65 years or older, suggesting that the spatial configuration underserved elderly populations (Fig. 3a). A similar pattern was generally observed in DFW and HOU, although perhaps to a lesser degree since there were notable exceptions where the cooling centers overlapped elderly population clusters (Figs. 4a and 5a).
The cooling center locations in SA appeared to substantially overlap clusters of individuals aged 5 years or younger and individuals with a disability (Figs. 3b,c). Overlap was similarly observed between the cooling center locations and young and disabled populations in DFW (Figs. 4b,c). However, one notable exception was identified: the spatial configuration underserved individuals with a disability throughout the peripheral regions of the DFW MSA. The cooling centers in HOU exhibited a less problematic relationship with disability clusters as the locations appeared to overlap a large portion of these areas (Fig. 5c). However, a lower degree of overlap was observed between the HOU cooling centers and areas of individuals aged 5 or younger, which suggests that younger populations were perhaps underserved (Fig. 5b).
Overall, the cooling center density surfaces and vulnerability hotspot cluster analyses highlighted a complex landscape where cooling centers were collocated with certain vulnerable populations but not others. Additionally, the specific vulnerable groups disadvantaged by the locations varied between the study MSAs. The spatial mismatches suggest that individuals with a disability warrant additional attention in DFW, while the youth and elderly would benefit from greater prioritization in HOU and SA, respectively. Although the insights provided by the visual overlays were helpful, more robust quantitative analysis of these relationships was explored via spatial lag regression.
d. Spatial lag regression
The SA MSA spatial lag regression model (Table 4) produced statistically significant results for two of the nine explanatory variables: aged at least 65 years and African American population. When interpreting the coefficient direction, positive relationships were considered problematic since they indicated that larger populations of vulnerable individuals were located at greater distances from a cooling center. Being aged at least 65 years was significantly positively associated with distance to the nearest cooling center, suggesting that elderly individuals in the SA MSA had limited cooling center access. This positive relationship aligned with the results from the hotspot cluster analysis, which highlighted a clear visual disconnect between clusters of individuals aged at least 65 years and the cooling center locations. It also supports previous work that highlighted the ineffectiveness of centralized cooling centers in reaching at-risk seniors (Bernard and McGeehin 2004). Conversely, the African American population exhibited a negative association, indicating that they were more proximate to cooling centers and likely benefited from greater access. A similar relationship was also identified for African American populations in Portland, Oregon (Voelkel et al. 2018).
Spatial lag regression results with distance to the nearest cooling center as the dependent variable for the SA MSA and Bexar County. Significant variables with p values less than 0.1 and less than 0.05 are indicated by italics and boldface type, respectively.
The DFW MSA spatial lag regression model (Table 5) produced statistically significant results for only the disability variable. Disability exhibited a positive relationship with distance to the nearest cooling center, which suggests that individuals with a disability were located at greater distances from cooling centers and were underserved by existing locations. The spatial regression results again supported the hotspot cluster analysis that identified a lack of overlap between individuals with a disability and the cooling center locations. Additionally, the relationship between distance to the nearest cooling center and no vehicle access was marginally significant (p = 0.06) and negative, indicating that housing units with no access to a personal automobile generally benefited from greater proximity to cooling centers.
Spatial lag regression results with distance to the nearest cooling center as the dependent variable for the DFW MSA, Dallas County, and Tarrant County. Significant variables with p values less than 0.1 and less than 0.05 are indicated by italics and boldface type, respectively.
Similar to the DFW MSA results, the HOU MSA spatial lag regression model (Table 6) produced statistically significant results for one variable: aged 5 years or younger. The association was negative, indicating that children in this age group were generally closer to cooling centers and likely had greater access relative to other population groups. Overall, the general absence of numerous significant associations between the socioeconomic indicators of vulnerability and proximity to a cooling center in the three MSA spatial regression models suggests that the cooling center locations throughout the Texas Triangle neither uniquely favored nor penalized the populations they typically aim to serve. While this indicates that cooling centers were generally not inequitably distributed throughout the MSAs, it also showcases how the emergent nature of the “networks” fails to comprehensively emphasize proximity to vulnerable populations.
Spatial lag regression results with distance to the nearest cooling center as the dependent variable for the HOU MSA and Harris County. Significant variables with p values less than 0.1 and less than 0.05 are indicated by italics and boldface type, respectively.
Because the cooling center locations were often confined to the urbanized cores of the MSAs, spatial lag regression was also performed for the core counties of each MSA. The Bexar County spatial regression model (Table 4) revealed a significant negative relationship between distance to the nearest cooling center and the percentage of the population that is Hispanic. Additionally, a marginally significant (p = 0.05) negative relationship was observed for the percentage of the population with income below the federal poverty level. Both relationships suggest that these two specific vulnerable communities within Bexar County were located closer to cooling centers, which is perhaps fortuitous, given the typical constraints governing cooling center site selection.
The analysis of the DFW MSA core considered Dallas and Tarrant counties individually. The Dallas County model produced no significant relationships, while the Tarrant County spatial regression model yielded two explanatory variables of significance: limited English-speaking households and the percentage of the population that is African American (Table 5). Both variables exhibited a negative association with distance to the nearest cooling center, suggesting that these two vulnerable groups in Tarrant County were generally closer to cooling centers. Similar to the Dallas County model, the Harris County spatial regression model (Table 6) produced no statistically significant relationships, with the exception of the percentage of the population that is African American, which had a marginally significant (p = 0.05) negative association. Unlike the results at the MSA scale, no significant positive relationships, which were indicative of inequitable cooling center distributions, were observed when considering only the core urban counties. However, while a limited number of vulnerable groups were located significantly closer to cooling centers, the lack of numerous negative significant relationships between the indicators of vulnerability and cooling center proximity suggests that cooling center locations failed to consistently prioritize all vulnerable population groups.
5. Conclusions
By utilizing spatial analysis techniques to analyze cooling center locations throughout the Texas Triangle, this study uncovered complex relationships between cooling center proximity and vulnerable populations. The cooling centers were generally located in the urban cores of the MSAs (with HOU being a slight exception), which raised questions regarding the potential heat vulnerability of more suburban and rural portions of the metropolitan areas. Similar issues regarding the lack of rural accessibility to cooling centers were identified by Nayak et al. (2019) in an analysis of New York. Furthermore, the cluster analysis of the locations indicated that spatial redundancies might be present in the current configurations due to potentially overlapping service areas. However, it is possible that the observed clustering may be intentional to service high-demand areas. In terms of cooling center accessibility, a small percentage of census tracts benefited from walkable access to a cooling center (1%–3%), which aligned with the analysis of Fraser et al. (2017). Accessibility improved notably when traveling via automobile and when public transit stops were considered. Improving walkability to cooling centers warrants additional attention because the assumption that heat-vulnerable residents own or have frequent and reliable access to an automobile and/or can easily navigate public transit networks is inherently problematic.
The spatial regression analysis indicated that Texas Triangle cooling center locations did not consistently marginalize or favor heat-vulnerable populations. This might be attributable to the emergent nature of cooling centers, as they typically rely on existing public buildings. However, there were several notable exceptions where significant positive associations were observed between vulnerable populations and distance to the nearest cooling center. Elderly individuals in SA, those with a disability in DFW, and children aged 5 years or younger in HOU all appeared to be at a disadvantage regarding cooling center access. Conversely, when considering solely the urban core counties of the MSAs, several significant negative relationships, indicating that vulnerable populations were closer to cooling centers, were observed. This nuance highlights the importance of considering scale when analyzing cooling center proximity to vulnerable communities.
Overall, since vulnerable populations were not consistently more proximate to cooling centers, greater intentionality and explicitly considering the locations of vulnerable communities could help optimize the spatial distribution of cooling centers and better align them with the areas of greatest need. This will likely require a broader strategic approach that shifts away from cooling center locations being determined primarily based upon convenience (i.e., existing publicly available facilities). Of course, additional research regarding individualized barriers to accessing cooling centers throughout Texas would be beneficial. This would likely involve studies utilizing surveys and interviews that enable a more detailed understanding of the characteristics of Texas Triangle cooling centers (e.g., use, capacity) and their visitors. Research of this nature could help further evaluate the degree to which cooling centers are meeting current needs and what potential changes could be most beneficial. Additionally, further refinement of the accessibility measures utilized in this study, such as evaluating different distance thresholds considered to be walkable, might provide additional insights that could inform cooling center planning. Despite these avenues of future research, this study highlights that the accessibility of cooling centers could be improved for certain vulnerable populations, which is critically important given the increasing frequency of extreme heat events throughout the Texas Triangle (Deng et al. 2018).
Acknowledgments.
The authors thank the reviewers and the editor for their constructive feedback that helped to improve the quality of the paper.
Data availability statement.
The socioeconomic data analyzed in this study were from the 2018 American Community Survey 5-yr estimates provided by the U.S. Census Bureau and are accessible online (https://data.census.gov/). The cooling center location data were gathered from local city government websites as well as news articles in the spring of 2021. The full list of cooling centers included in the analysis is provided in Tables S1–S3 in the online supplemental material.
REFERENCES
Allen, M., J. Hoffman, J. L. Whytlaw, and N. Hutton, 2022: Assessing Virginia cooling centers as a heat mitigation strategy. J. Emerg. Manage., 20, 205–224, https;//doi.org/10.5055/jem.0671.
Bao, J., X. Li, and C. Yu, 2015: The construction and validation of the heat vulnerability index, a review. Int. J. Environ. Res. Public Health, 12, 7220–7234, https://doi.org/10.3390/ijerph120707220.
Bedi, N., Q. Adams, J. Hess, and G. Wellenius, 2022: The role of cooling centers in protecting vulnerable individuals due to extreme heat. Epidemiology, 33, 611–615, https://doi.org/10.1097/EDE.0000000000001503.
Bedsworth, L., 2009: Preparing for climate change: A perspective from local public health officers in California. Environ. Health Perspect., 117, 617–623, https://doi.org/10.1289/ehp.0800114.
Berisha, V., and Coauthors, 2017: Assessing adaptation strategies for extreme heat: A public health evaluation of cooling centers in Maricopa County, Arizona. Wea. Climate Soc., 9, 71–80, https://doi.org/10.1175/WCAS-D-16-0033.1.
Bernard, S., and M. McGeehin, 2004: Municipal heat wave response plans. Amer. J. Public Health, 94, 1520–1522, https://doi.org/10.2105/AJPH.94.9.1520.
Betancourt, J. R., A. R. Green, J. E. Carrillo, and O. Ananeh-Firempong II, 2003: Defining cultural competence: A practical framework for addressing racial/ethnic disparities in health and health care. Public Health Rep., 118, 293–302, https://doi.org/10.1016/S0033-3549(04)50253-4.
Bouchama, A., M. Dehbi, G. Mohamed, F. Matthies, M. Shoukri, and B. Menne, 2007: Prognostic factors in heat wave–related deaths: A meta-analysis. Arch. Intern. Med., 167, 2170–2176, https://doi.org/10.1001/archinte.167.20.ira70009.
Bradford, K., L. Abrahams, M. Hegglin, and K. Klima, 2015: A heat vulnerability index and adaptation solutions for Pittsburgh, Pennsylvania. Environ. Sci. Technol., 49, 11 303–11 311, https://doi.org/10.1021/acs.est.5b03127.
Braveman, P., S. Egerter, and D. Williams, 2011: The social determinants of health: Coming of age. Annu. Rev. Public Health, 32, 381–398, https://doi.org/10.1146/annurev-publhealth-031210-101218.
City of Dallas, 2020: Dallas comprehensive environmental and climate action plan. City of Dallas Doc., 122 pp., https://www.dallasclimateaction.com/_files/ugd/349b65_e4f9a262cebf41258fd4343d9af0504f.pdf.
City of San Antonio, 2019: SA Climate Ready: A pathway from climate action and adaptation. City of San Antonio 2019 Rep., 92 pp., https://www.sanantonio.gov/Portals/0/Files/Sustainability/SAClimateReady/SACRReportOctober2019.pdf.
Cutter, S., B. Boruff, and W. Shirley, 2003: Social vulnerability to environmental hazards. Soc. Sci. Quart., 84, 242–261, https://doi.org/10.1111/1540-6237.8402002.
Dahl, K., R. Licker, J. T. Abatzoglou, and J. Declet-Barreto, 2019: Increased frequency of and population exposure to extreme heat index days in the United States during the 21st century. Environ. Res. Commun., 1, 075002, https://doi.org/10.1088/2515-7620/ab27cf.
Deng, K., M. Ting, S. Yang, and Y. Tan, 2018: Increased frequency of summer extreme heat waves over Texas area tied to the amplification of Pacific zonal SST gradient. J. Climate, 31, 5629–5647, https://doi.org/10.1175/JCLI-D-17-0554.1.
Fechter-Leggett, E., A. Vaidyanathan, and E. Choudhary, 2016: Heat stress illness emergency department visits in National Environmental Public Health Tracking states, 2005–2010. J. Community Health, 41, 57–69, https://doi.org/10.1007/s10900-015-0064-7.
Fraser, A. M., M. V. Chester, D. Eisenman, D. M. Hondula, S. S. Pincetl, P. English, and E. Bondank, 2017: Household accessibility to heat refuges: Residential air conditioning, public cooled space, and walkability. Environ. Plann., 44B, 1036–1055, https://doi.org/10.1177/0265813516657342.
Gaskin, C., D. Taylor, S. Kinnear, J. Mann, W. Hillman, and M. Moran, 2017: Factors associated with the climate change vulnerability and the adaptive capacity of people with disability: A systematic review. Wea. Climate Soc., 9, 801–814, https://doi.org/10.1175/WCAS-D-16-0126.1.
Getis, A., and J. K. Ord, 1992: The analysis of spatial association by use of distance statistics. Geogr. Anal., 24, 189–206, https://doi.org/10.1111/j.1538-4632.1992.tb00261.x.
Guerra, E., R. Cervero, and D. Tischler, 2012: Half-mile circle: Does it best represent transit station catchments? Transp. Res. Rec., 2276, 101–109, https://doi.org/10.3141/2276-12.
Hansen, A., L. Bi, A. Saniotis, and M. Nitschke, 2013: Vulnerability to extreme heat and climate change: Is ethnicity a factor? Global Health Action, 6, 21364, https://doi.org/10.3402/gha.v6i0.21364.
Hayden, M. H., and Coauthors, 2017: Adaptative capacity to extreme heat: Results from a household survey in Houston, Texas. Wea. Climate Soc., 9, 787–799, https://doi.org/10.1175/WCAS-D-16-0125.1.
Hoffman, J. S., V. Shandas, and N. Pendleton, 2020: The effects of historical housing policies on resident exposure to intra-urban heat: A study of 108 US urban areas. Climate, 8, 12, https://doi.org/10.3390/cli8010012.
Jagai, J. S., E. Grossman, L. Navon, A. Sambanis, and S. Dorevitch, 2017: Hospitalizations for heat-stress illness varies between rural and urban areas: An analysis of Illinois data, 1987–2014. Environ. Health, 16, 38, https://doi.org/10.1186/s12940-017-0245-1.
Johnson, D., A. Stanforth, V. Lulla, and G. Luber, 2012: Developing an applied extreme heat vulnerability index utilizing socioeconomic and environmental data. Appl. Geogr., 35, 23–31, https://doi.org/10.1016/j.apgeog.2012.04.006.
Kenny, G. P., J. Yardley, C. Brown, R. J. Sigal, and O. Jay, 2010: Heat stress in older individuals and patients with common chronic diseases. Can. Med. Assoc. J., 182, 1053–1060, https://doi.org/10.1503/cmaj.081050.
Kim, K., J. Jung, C. Schollaert, and J. Spector, 2021: A comparative assessment of cooling center preparedness across twenty-five U.S. cities. Int. J. Environ. Res. Public Health, 18, 4801, https://doi.org/10.3390/ijerph18094801.
Lehnert, E., G. Wilt, B. Flanagan, and E. Hallisey, 2020: Spatial exploration of the CDC’s Social Vulnerability Index and heat-related health outcomes in Georgia. Int. J. Disaster Risk Reduct., 46, 101517, https://doi.org/10.1016/j.ijdrr.2020.101517.
Li, D., G. Newman, B. Wilson, Y. Zhang, and R. D. Brown, 2022: Modeling the relationships between historical redlining, urban heat, and heat-related emergency department visits: An examination of 11 Texas cities. Environ. Plann., 49B, 933–952, https://doi.org/10.1177/23998083211039854.
Luber, G., and M. McGeehin, 2008: Climate change and extreme heat events. Amer. J. Prev. Med., 35, 429–435, https://doi.org/10.1016/j.amepre.2008.08.021.
Mallen, E., and Coauthors, 2022: Extreme heat exposure: Access and barriers to cooling centers–Maricopa and Yuma Counties, Arizona, 2010–2020. Morb. Mortal. Wkly. Rep., 71, 781–785, https://doi.org/10.15585/mmwr.mm7124a1.
Marsha, A., S. R. Sain, M. J. Heaton, A. J. Monaghan, and O. V. Wilhelmi, 2018: Influences of climatic and population changes on heat-related mortality in Houston, Texas, USA. Climatic Change, 146, 471–485, https://doi.org/10.1007/s10584-016-1775-1.
Nayak, S., S. Lin, S. C. Sheridan, Y. Lu, N. Graber, M. Primeau, C. J. Rafferty, and S.-A. Hwang, 2017: Surveying local health departments and county emergency management offices on cooling centers as a heat adaptation resource in New York State. J. Community Health, 42, 43–50, https://doi.org/10.1007/s10900-016-0224-4.
Nayak, S., and Coauthors, 2019: Accessibility of cooling centers to heat-vulnerable populations in New York State. J. Transp. Health, 14, 100563, https://doi.org/10.1016/j.jth.2019.05.002.
Ord, J. K., and A. Getis, 1995: Local spatial autocorrelation statistics: Distributional issues and an application. Geogr. Anal., 27, 286–306, https://doi.org/10.1111/j.1538-4632.1995.tb00912.x.
Ostro, B., S. Rauch, R. Green, B. Malig, and R. Basu, 2010: The effects of temperature and use of air conditioning on hospitalizations. Amer. J. Epidemiol., 172, 1053–1061, https://doi.org/10.1093/aje/kwq231.
Raphael, D., 2006: Social determinants of health: Present status, unanswered questions, and future directions. Int. J. Health Serv., 36, 651–677, https://doi.org/10.2190/3MW4-1EK3-DGRQ-2CRF.
Sampson, N., and Coauthors, 2013: Staying cool in a changing climate: Reaching vulnerable populations during heat events. Global Environ. Change, 23, 475–484, https://doi.org/10.1016/j.gloenvcha.2012.12.011.
Sharif, H., 2018: Climate projections for the city of San Antonio. City of San Antonio, 17 pp., https://www.sanantonio.gov/Portals/0/Files/Sustainability/SAClimateReady/San-Antonio-Climate-Projections.pdf.
Stanberry, L. R., M. C. Thomson, and W. James, 2018: Prioritizing the needs of children in a changing climate. PLOS Med., 15, e1002627, https://doi.org/10.1371/journal.pmed.1002627.
Uejio, C., O. V. Wilhelmi, J. S. Golden, D. M. Mills, S. P. Gulino, and J. P. Samenow, 2011: Intra-urban societal vulnerability to extreme heat: The role of heat exposure and the built environment, socioeconomics, and neighborhood stability. Health Place, 17, 498–507, https://doi.org/10.1016/j.healthplace.2010.12.005.
U.S. Census Bureau, 1994: Geographic areas reference manual. U.S. Department of Commerce, 45 pp.
U.S. Census Bureau, 2018: American Community Survey. Accessed 1 February 2021, https://data.census.gov/advanced?g=040XX00US48$1400000&y=2018&d=ACS%205-Year%20Estimates%20Detailed%20Tables.
U.S. Census Bureau, 2021: American Housing Survey. Accessed 1 March 2023, https://www.census.gov/programs-surveys/ahs/data/interactive/ahstablecreator.html.
Vaidyanathan, A., J. Malilay, P. Schramm, and S. Saha, 2020: Heat-related deaths – United States, 2004–2018. Morb. Mortal. Wkly. Rep., 69, 729–734, https://doi.org/10.15585/mmwr.mm6924a1.
Vandentorren, S., and Coauthors, 2006: August 2003 heat wave in France: Risk factors for death of elderly people living at home. Eur. J. Public Health, 16, 583–591, https://doi.org/10.1093/eurpub/ckl063.
Voelkel, J., D. Hellman, R. Sakuma, and V. Shandas, 2018: Assessing vulnerability to urban heat: A study of disproportionate heat exposure and access to refuge by socio-demographic status in Portland, Oregon. Int. J. Environ. Res. Public Health, 15, 640, https://doi.org/10.3390/ijerph15040640.
Weinberger, K. R., D. Harris, K. R. Spangler, A. Zanobetti, and G. Wellenius, 2020: Estimating the number of excess deaths attributable to heat in 297 United States counties. Environ. Epidemiol., 4, e096, https://doi.org/10.1097/EE9.0000000000000096.
Whitman, S., G. Good, E. R. Donoghue, N. Benbow, W. Shou, and S. Mou, 1997: Mortality in Chicago attributed to the July 1995 heat wave. Amer. J. Public Health, 87, 1515–1518, https://doi.org/10.2105/AJPH.87.9.1515.
Widerynski, S., and Coauthors, 2017: The use of cooling centers to prevent heat-related illness: Summary of evidence and strategies for implementation. Centers for Disease Control and Prevention, 36 pp., https://www.cdc.gov/climateandhealth/docs/UseOfCoolingCenters.pdf.
Yardley, J., R. Sigal, and G. Kenny, 2011: Heat health planning: The importance of social and community factors. Global Environ. Change, 21, 670–679, https://doi.org/10.1016/j.gloenvcha.2010.11.010.
